Increase cache hit ratio for push

We don't update the cache in all code paths, which causes subsequent
calls to miss the cache

Add more metrics to push rule evaluation

Push: Don't acquire lock unless necessary

Use less DB for device list handling in sync

encode sync-response statically

Don't compute push actions for backfilled events

Add a frontend proxy

Perf: Don't filter events for push

We know the users are joined and we can explicitly check for if they are
ignoring the user, so lets do that.

Fix bug where pusherpool didn't start and broke some rooms

Since we didn't instansiate the PusherPool at start time it could fail
at run time, which it did for some users.

This may or may not fix things for those users, but it should happen at
start time and stop the server from starting.

Fix up user_ip replication commands

Separate federation servlet into different lists

Fix caching error in the push evaluator

Initialising `result` to `{}` in the parameters meant that every call to
_flatten_dict used the *same* target dictionary.

I'm hopeful this will fix https://github.com/matrix-org/synapse/issues/2270,
but I suspect it won't. (This code seems to have been here since forever,
unlike the bug, and I don't really think it explains the observed
behaviour). Still, it makes it hard to investigate the problem.

Increase default cache size

Improve the error handling for bad invites received over federation