Error refactor

See merge request soprani.ca/sgx-bwmsgsv2!13

Initial test framework and some tests

See merge request soprani.ca/sgx-bwmsgsv2!12

Reply to IBR remove request

See merge request soprani.ca/sgx-bwmsgsv2!11

Otherwise caller times out waiting for it

Need Redis 7 for NX and GET together

See merge request soprani.ca/sgx-bwmsgsv2!10

Instrument for sentry

See merge request soprani.ca/sgx-bwmsgsv2!9

Unregister when asked

See merge request soprani.ca/sgx-bwmsgsv2!7

v1 is dead, remove hacks

See merge request soprani.ca/sgx-bwmsgsv2!8

Keeps all knowledge about redis key structure, etc, in one place.

Not clean yet, but at least it runs

Use new em_promise.rb gem

See merge request soprani.ca/sgx-bwmsgsv2!2

Remove unused redis usage data

See merge request soprani.ca/sgx-bwmsgsv2!6

Simplify iq feature-not-implemented

See merge request soprani.ca/sgx-bwmsgsv2!5

This redis key is not used anymore

Send to the correct target, do not assume cheogram

See merge request soprani.ca/sgx-bwmsgsv2!4

Backport from prod hotfix

Log the errorCode and description for deliver-failure

See merge request soprani.ca/sgx-bwmsgsv2!3

Let them see the human readable failure description

deliveryCode and deliveryDescription were the v1 equivalents, but this is v2

Specifically https://gitlab.com/ossguy/sgx-catapult/commit/583f98b is
copied here.  While there was another commit in that merge request (at
https://gitlab.com/ossguy/sgx-catapult/merge_requests/20), that commit
(https://gitlab.com/ossguy/sgx-catapult/commit/8f9e588) was already
included in 9d556e1 so we don't need to copy the entire merge request
here, only the single commit in that merge request that included the
"Allow EMPromise.all to accept both Promise and Deferable" part.

Note that this is required in sgx-bwmsgsv2 since jmp-fwdcalls uses it
and jmp-fwdcalls must be able to run with either sgx-catapult or
sgx-bwmsgsv2.  As an example, the following commit and MR require it:
https://gitlab.com/ossguy/jmp-fwdcalls/commit/405ef3c and
https://gitlab.com/ossguy/jmp-fwdcalls/merge_requests/12 .

We are doing the copy here since we haven't properly refactored the
common parts of sgx-catapult and sgx-bwmsgsv2 so they can, for
example, share one em_promise.rb file.

See merge request !1 for the discussion and details behind the merge.

The metrics changes in .rubocop.yml definitely need to be reset to defaults, but
the required refactors are too big for this change alone so for now they were
increased to make it pass.

Other disabled rules may want to be revisited in the future also.

At least this passes its own ruleset now, and can run on sourcehut CI for every
commit.

This fix is ported in from sgx-catapult, where we made the fix in
https://gitlab.com/ossguy/sgx-catapult/commit/c8fd695 - it is fairly
simple, but also fairly important.  Here is the description (with
edits made to show the corresponding commit IDs in this repo):

Sometimes HTTP requests to the port that sgx-bwmsgsv2 is running on
are slightly less naive - for example, with a path of:

  /web-meetme/conf_cdr.php?bookId=1

In this case params is indeed non-empty so we need to also catch this
case.  To be extra safe, we do so by rejecting both non-POST and
non-"root" requests.  Hopefully this will suffice for the situations
we care about, though of course the better solution would be to do
actual request validation.  We'll save that for later.

As with eddafea we have the same error/crash without this fix (if we
got an HTTP request of the above form):

  Shutting down gateway due to exception 013: no implicit conversion of nil into String

A copy of https://gitlab.com/ossguy/sgx-catapult/commit/a887c61 :

This is not 100% correct, but it's close enough for our purposes and
we left a TODO in case someone wants to fix it properly in the future.
This fix will at least prevent feedback loops of the type we had
previously with components like Cheogram.

This fix is ported in from sgx-catapult, where we made the fix in
https://gitlab.com/ossguy/sgx-catapult/commit/300def9 - it is fairly
simple, but also fairly important.  Here is the description:

Without this fix, one will get the following error (and subsequent
crash) if the port that sgx-catapult is running on receives a naive
HTTP request, such as a "GET /" (from Wget or curl or similar):

  Shutting down gateway due to exception 013: no implicit conversion of nil into String

This is a cross-merge of the fix done to sgx-catapult in
https://gitlab.com/ossguy/sgx-catapult/commit/89598a8 - it has no
effect in sgx-bwmsgsv2 for now, but will help once we cross-merge the
blocking code (to be done later).

We just have one fix left until we otherwise have fix-/feature-parity
with sgx-catapult.

This is a cross-merge of what we did in sgx-catapult already, in
https://gitlab.com/ossguy/sgx-catapult/commit/f95aab3 - see
https://gitlab.com/ossguy/sgx-catapult/merge_requests/19 for further
details.  The merge is not exact, as there has been some drift in
sgx-bwmsgsv2 since this was done in sgx-catapult, so indentation and
some s/next/return/ caused it to be a bit off.  But it should be fine.

This effectively removes the option to turn off MMS-on-OBB, and since
that has already been removed in sgx-catapult (per above), we can now
remove it from jmp-acct_bot as there are no further users (to be done
in a future commit).

With this commit we are getting close to feature-/fix-parity with
sgx-catapult, aside from the message blocking feature (which we won't
need to merge for a while yet).

We run sgx-bwmsgsv2 and sgx-catapult off the same database, which
means that mere existence in the database does not mean that we can
pass on the message directly, as they could be on a different system
(i.e. sgx-catapult instead of sgx-bwmsgsv2) and Cheogram will become
unhappy if we try to pass on messages in this way (plus it's wrong).

As a result, we need to make the check more precise, and specifically
determine that the destination user is a user of sgx-bwmsgsv2, which
we do by checking for a distinctive characteristic of the credentials.
This should ideally be done in a better way, but since we just have
the two systems for now, it is easy to do this substring check, as all
credentials follow this general pattern.

We know this commit works because we've been running it in production
for a couple months now (* sigh *) and will be making a similar commit
to sgx-catapult in the near future (as it naturally requires the same
sort of change in order to work properly, per above).

While this localhost hardcoding would work fine if sgx-bwmsgsv2 was
run on its own, it usually isn't.  Instead, it is run as a plugover
(see https://gitlab.com/ossguy/jmp-fwdcalls/commit/08d3acc for
details), and because the voice endpoint that jmp-fwdcalls responds to
is public (so our carrier can deliver voice-related requests directly
to it), we need the whole endpoint to be public, even though the
requests delivered to sgx-bwmsgsv2 don't require that.

Ah, the idealism of 91ff289 could simply not be left unsullied.  We
though the V2 SGX would only need to deal with V2 media, but alas it
must deal with V1 media too, since we want it to work with the V1
jmp-fwdcalls.  In particular this means that send_media() must handle
the V1 media that is sent by jmp-fwdcalls (voicemail recording URLs).

This is roughly the counterpart of the fix we made at
https://gitlab.com/ossguy/sgx-catapult/commit/7dfe149683d31738076b075e9815643194d81684
- check the prefix and set the user accordingly (instead of checking
the user to set the prefix accordingly).

While we should probably document that deployment requires this extra
user, the warning in the code should serve the small number of people
who actually care well enough.  We hopefully won't need to run this
V1 + V2 monstrosity for very long.

With this fix we now fully support voicemails received by V2 users who
happen to still be using the V1 API for voice (which happens to be
everyone right now).  We thought this was all done as of
https://gitlab.com/ossguy/jmp-fwdcalls/commit/a4ed555e2dd16fb342f33ea7ee5675553baa39ec
but we needed the commit here to make voicemail recordings work.  I
suppose the jmp-fwdcalls commit message is still correct, though - no
additional work was needed in jmp-fwdcalls to fix the voicemail
recordings: only the fix here was required.

Since we ripped out support for Jingle File Transfer in 60832f5 we
need to figure out the best way to provide MMS by default to users of
this SGX (the V2 SGX).  Should we just turn it on by default for all
users?  We could, but we'd have to let anyone migrating from V1 to V2
know about this change.  And maybe there should be a different default
option?  In any case, it still needs to be decided so add a TODO here.

This is a continuation of the work started in 90a21bc, which prepares
the SGX for integration with jmp-fwdcalls and its V1 dependence.  Left
out from that previous commit was a mention of where we first did the
V1->V2 switch in this SGX.  For the URL code of important to the
call_catapult() method, this was in a9237ee.

Back to this commit, we have paved the way for jmp-fwdcalls to pass us
a config file containing the V1 credentials, which we can then use
when we detect a V1 path in call_catapult() and convert accordingly.

Until Bandwidth supports SIP endpoints in their V2 voice API, we need
to use their V1 voice API.  However, to use the V1 voice API alongside
the V2 messages API, we need to be a bit clever.  In particular, we
need to detect when we're getting a V1 call and adjust the URL
accordingly.

The hack implemented here to accomplish this allows us to make very
minimal changes to jmp-fwdcalls (letting it continue to use an
unmodified call_catapult() invocation) while still permitting it to
work with the V2 SGX (this one) as well as the V1 SGX.  As noted in
this commit's comments, jmp-fwdcalls will need to pass us some V1
credentials to use (since the V2 user won't have these available in
its creds list).  But that should be the only added complexity here.

We added group picture message receiving support in d89a1b7, but the
API introduced there (as part of the send_media method) was not ideal
due to the caller having to worry about the deep clone.  So instead we
will move the .copy() into send_media() so that it worries about the
deep clone and the caller doesn't have to care.