The `@cached` decorator on `KeyStore._get_server_verify_key` was missing
its `num_args` parameter, which meant that it was returning the wrong key for
any server which had more than one recorded key.

By way of a fix, change the default for `num_args` to be *all* arguments. To
implement that, factor out a common base class for `CacheDescriptor` and `CacheListDescriptor`.

Don't send the full event json over replication

Add some debug to help diagnose weird federation issue

Logcontext docs

Apparently the github RST renderer doesn't like Note blocks.

A few fixes to logcontext things

Avoid resetting state on rejected events

Fix a bug in ``logcontext.preserve_fn`` which made it leak context into the
reactor, and add a test for it.

Also, get rid of ``logcontext.reset_context_after_deferred``, which tried to do
the same thing but had its own, different, set of bugs.

This fixes a class of 'Unexpected logcontext' messages, which were happening
because the logcontext was somewhat arbitrarily swapping between the sentinel
and the `run` logcontext.

background_updates was using `call_later` in a way that leaked the logcontext
into the reactor.

We could have rewritten it to do it properly, but given that we weren't using
the fancier facilities provided by `call_later`, we might as well just use
`async.sleep`, which does the logcontext stuff properly.

When we get a rejected event, give it the same state_group as its prev_event,
rather than no state_group at all.

This should fix https://github.com/matrix-org/synapse/issues/1935.

A few non-functional changes:

* A bunch of docstrings to document types
* Split `EventsStore._persist_events_txn` up a bit. Hopefully it's a bit more
  readable.
* Rephrase `EventFederationStore._update_min_depth_for_room_txn` to avoid
  mind-bending conditional.
* Rephrase rejected/outlier conditional in `_update_outliers_txn` to avoid
  mind-bending conditional.

Queue up federation PDUs while a room join is in progress

Implement no op for room stream in sync

This just takes the existing `room_queues` logic and moves it out to
`on_receive_pdu` instead of `_process_received_pdu`, which ensures that we
don't start trying to fetch prev_events and whathaveyou until the join has
completed.

Add fallback to last_active_ts if it beats the last sync time on a presence timeout.

Format presence events on the edges instead of reformatting them multiple times

Handle PartialDownloadError in CAS login

Fix assertion to stop transaction queue getting wedged

... and update some docstrings to correctly reflect the types being used.

get_new_device_msgs_for_remote can return a long under some circumstances,
which was being stored in last_device_list_stream_id_by_dest, and was then
upsetting things on the next loop.

Add some metrics on notifier