Not only is it allowed, but actually there's a ton of empty notes. Every
monthly account activation is missing a note, so practically every user
that has transactions will have empty notes.

Some automated processes don't care for our menu, and rather than
crashing it should just consider our process to be done.

* tx-list:
  Admin Command Menu + Admin Financial View
  Transactions List
  Customer Financials
  Telephone Link

I've added a new command to show financial information about the user.
But more importantly I've added an infinite admin subsystem which allows
me to go into a user and then run multiple commands on them.

For now it's just these two info commands, but in the future I'd like to
add mutative commands here.

Finally, since I sometimes look up multiple users in a pretty short
timeframe I made the menu open and if I put something into the list that
doesn't parse as one of the actions it instead switches the current user
so I don't have to do "cancel", "customer", and then put the next one
in.

I can just have the session open and put stuff in as needed.

Using the new Financials object a user can list their transaction list
We only show the user the date, amount, and method.

I've pulled out information about payment methods and bitcoin into its
own thing. It's kind of a repository, except that it only exposes
fetchers and doesn't load anything generally.

There's also a few new methods here that aren't used yet, but will be
shortly.

Sometimes I need to go there to look something up or change a setting,
and every time I do I have to navigate through the UI to a number I
already know.

This should make that faster.

* state-search:
  Finish state search and add test
  Add state search - for granularity & jmp-register

Note that this is not quite complete yet, since we need to pull in the
STATE_MAP stuff used elsewhere to make it work properly.  But it will
at least work for every state/province/territory that is not Quebec
for now, which is way better than nothing.  Quebec users will get the
same fallbacks they had before.

If the SGX fetch fails, this is a problem.  Even in the original case this was
written for (registration) we have seen strange issues that can come up when the
code assumes that an SGX error means they are unregistered.  They may or may not
be unregistered, and we simply don't know.  So we should let the error bubble
through and be handled (if possible) upstream.

The to in leg2 is not the phone number, but the fwd, which is not useful for
looking up customer.

Doesn't affect any current code paths, but should be correct in case we ever
start using it in the inbound case.

Also from does not have to be a formatted phone number, can be a string like
Anonymous, etc.

For use from Asterisk

There is an edge case where a customer might not have auto-topped up yet but
they don't have enough balance for this call, so try to charge their card first
before telling them the call is a no go.

If cannot find an acceptable rate for the number, cannot call.
If balance is too low, cannot call.
If too close to limit, warn.
Else, call.

This will allow us to switch asterisk to the new logic before rolling out the
limits.  Asterisk will then support the limits once sgx-jmp begins instructing
it about them.

Introduce objects representing the different kind of query keys that can be used
to look up a customer.  These keys know how to use themselves to get
the (customer_id, jid) pair that is needed for an actual repo lookup.  All
find_by_* methods are now just trivial wrappers over these objects that pass on
to find_inner.

This enables the easy creation of a new helper: QueryKey.for, used as
find_by_format which allows passing an xmpp: URI, or a properly formatted tel,
or a customer id, and will find by the correct one.  This could cause a problem
if a customer id were to be a valid xmpp: URI or a formatted tel, so this method
does assume not using these formats for any customer id (which would be
ill-advised in any case).

Generic error message to contact support.  No current customer will hit this
limit based on historical usage.

We've started getting lax about vetting the warnings because there are too many.
Raise the threshold for a warning.

This is based on a hotfix in prod. In this case someone appeared to have
talked to an @jmp.chat address directly. In normal circumstances this
shouldn't happen, but it's not technically disallowed, so we probably
shouldn't just crash.