Instead of having a second method to fetch channel-binding from the
TlsStream, do it directly in the connect() method, since after that we
don’t have enough information to fetch it any longer when using ktls.

This disables some tests, but those were controversial anyway.

I was thinking about replacing the Feature struct with a plain String,
what do you think about it?

The two remaining issues, which led to two ignored tests, are that
priority now gets always serialized, and that we don’t reject duplicated
identical xml:lang in statuses.

This was previously unimplemented due to how the enum was treated, now
that we moved the reason element to xso we can describe it properly and
thus implement everything it supports.

This causes one regression on xml:lang validation, but we can live with
that until xso implements such verification.

Without this, it'd panic, which is "not ideal", as they say.

XEP-0045 doesn’t specify what to do in the case the actor contains both
a @jid and a @nick, so let’s not encode anything special about that
here.

In addition this converted it to use xso.

Supersedes !196.

Fixes #88.

Oftentimes, Stanza is at hand anyway, so this saves some typing.

This allows to use any serialisable type. The advantage is that moves
and clones are avoided (which would otherwise be needed to construct
e.g. a XmppStreamElement from a Stanza or Message).

This is useful in situations where we don't want to care about the
specific type of the underlying transport.

Version 0.4.2 added a restriction on width and height to be in the
0..65535 range.  The media-type attribute has always been optional as
well.

This attribute is now a simple bool instead of an enum.

These bring basically nothing, so we can throw them away alongside their
generator macro.

Signed-off-by: Maxime “pep” Buquet <pep@bouah.net>

This also keeps all additional elements in the vCard, so they get
serialized back on the wire instead of being dropped.

This XEP is used to synchronize where each client has stopped reading a
conversation, so that we can e.g. stop displaying a notification when
the user has read this particular message on a different device.

Use `GenericArray::to_vec()` and `slice::to_vec()` directly.

Use `GenericArray::to_vec()` directly.

This tests each element we support, and also keep track of the sizes.

Additionally the copyright year was wrong.

Since Rust 1.81, std::error::Error has been moved to core::error::Error,
which we can use without depending on std.

We mostly had to import from alloc some structs that are part of the std
prelude, such as Vec and String.

All keys only have a single char, so avoid allocating a whole String for
them.  Also move to a BTreeMap as this is more efficient on such keys.

This drastically improve the debuggability of the network parts, by
moving the task of encrypting/decrypting TLS packets from userland to
the kernel.  This makes them appear in clear in strace as sendto() and
recvfrom().

I introduced a new tls-rust-ktls feature which depends on both rustls
and ktls, but isn’t enabled by default since it probably isn’t available
on every computer.  It requires the tls kernel module to be loaded,
which then can offload encryption/decryption to dedicated hardware if
available.

I have tested this change on Linux 6.11 on a rk3588.

It started breaking CI for some reason, probably a rustfmt update.