... to avoid locking the table for too long

Add a flag which makes the purger delete local events

it was a bit of a non-sequitur there

this thing takes ages and the only sign of any progress is the logs, so having
some logs is useful.

(beacause it deletes more than state)

In particular I hope this will help the pusher, which makes many requests to
sygnal, and is currently negotiating SSL for each one.

... which I introduced in #2785

better exception logging in callbackmetrics

when we fail to render a metric, give a clue as to which metric it was

Log room when doing state resolution

Mostly because it helps figure out what is prompting the resolution

Keep track of last access time for local media

fix SQL when searching all users

Split resolve_events into two functions

... so that the return type doesn't depend on the arg types

Track DB scheduling delay per-request

Sanity checking for user ids

Check the user_id passed to a couple of APIs for validity, to avoid
"IndexError: list index out of range" exception which looks scary and results
in a 500 rather than a more useful error.

Fixes #1432, among other things

Track db txn time in millisecs

Fix 'NoneType' object has no attribute 'writeHeaders'

rework runInteraction in terms of runConnection

Optimise LoggingContext creation and copying

Fix thumbnailing remote files

Avoid throwing a (harmless) exception when we try to write an error response to
an http request where the client has disconnected.

This comes up as a CRITICAL error in the logs which tends to mislead people
into thinking there's an actual problem

For each request, track the amount of time spent waiting for a db
connection. This entails adding it to the LoggingContext and we may as well add
metrics for it while we are passing.