those links are used for attachments, "enclosure" for files, "related" when it's other
kind of data, such as an external website (in this case the `external` key is set in the
attchment).
when a direct (not public) message is sent to AP peer, a mention should be present
otherwise the peer won't receive a notification in most AP implementation (notably
Mastodon).
When an `exceptions.NotFound` is received, log a warning message and return the
appropriate HTTP code instead of raising the default `exceptions.InternalError`.
OMEMO is not used in components so far, but the trigger is trying to request OMEMO PEP
nodes, which causes an error with virtual pubsub service of AP component.
XEP-0384 connection workflow can be very long when connecting after a while (due to
OLDMEMO migration notably), which can lead to bad UX.
It is not necessary to `await` it as `encrypt` and `decrypt` already wait for the plugin
to be initialized correctly, so the workflow is now run in background.
MAM retrieval can be long, and can be done after connection, message just need to be
sorted when being inserted (i.e. frontends must do insort).
To avoid blocking connection for too long and result in bad UX and timeout risk, one2one
MAM message are not retrieved in background.
in `_join_MAM`, `room.fully_joined` is called before retrieving the MAM archive, as the
process can be very long, and is not necessary to have the room working (message can be
received after being in the room, and added out of order). This avoid blocking the `join`
workflow for an extended time.
Some renaming and coroutine integrations.
Application may be long to start (e.g. a Docker app may have to download images first,
and even without the downloading, the starting could be long), which may lead to UI
blocking or bridge time out.
To prevent that, `start` is now returning immediately, and 2 new signals are used to
indicate when the application is started, of if something wrong happened.
`start` now returns initial app data, including exposed data without the computed exposed
data. The computed data must be retrieved after the app has been started.
Mastodon doesn't like when "@" is percent-encoded (because it doesn't do it, and it checks
that the URL matches the ID without taking care of percent-encoding).
When a `Delete` activity was received and the object was the emitting actor itself, the
signature checking was failing if the actor was unknown (due to the impossibility to
retrieve the actor public key, as it is no more accessible).
To avoid that, those notifications are ignored for now. In the future they should clean
the cache linked to this actor.
if something is going wrong during `checkSignature` (other than signature verification
failing), a `500 Internal Server Error` code is returned, and the connection is properly
closed.
the body is parsed immediately during a `POST` request: this avoids duplication of code,
and allows to check the body data before continuing (will be used to filter some requests
in a future patch).
