Fix a bug where the `get_joined_users` cache could be corrupted by custom
status events (or other state events with a state_key matching the user ID).

The bug was introduced by #2229, but has largely gone unnoticed since then.

Fixes #7099, #7373.

This backs out some of the validation for the client dictionary and logs if
this changes during a user interactive authentication session instead.

This is a cherry-pick of 1a1da60a (#7470)
to the release-v1.13.0 branch.

Be less strict about validation of UI authentication sessions during
registration to match client expecations.

Since we only use pubsub, the dbid is irrelevant.

Add dummy_events_threshold which allows configuring the number of forward extremities a room needs for Synapse to send forward extremities in it.

For the record, the reason we need this is as follows:

each RDATA command comes down the redis pipe as a subscription message. txredisapi as written needs at least three reactor ticks to read each subscription message from the tcp buffer. Hence, once the process gets loaded, it starts getting behind, and eventually redis knifes the connection. it then takes ages for the master to work its way through the backlog, before it reconnects again, during which any commands from any workers are dropped.

Speed up fetching device lists changes in sync.

Fix lost events on replication reconnection

... otherwise we can believe we're up to date when we're not.

Currently we copy `users_who_share_room` needlessly about three times,
which is expensive when the set is large (which it can easily be).

This caused `prev_state_ids` to be incorrect if the state event was not
replacing an existing state entry.

Fix MANIFEST.in