Add more useful logging when we block fetching events

Remove fallback from get_missing_events.

get_missing_events used to fallback to fetching the missing events
individually requesting from every server in the room, one by one.e

This could be unacceptably slow, possibly causing #1732

Use the new twisted logging framework.

Hopefully adding an observer to the new framework will avoid a memory
leak https://twistedmatrix.com/trac/ticket/8164

Wrap connections in an N minute timeout to ensure they get reaped correctly

The abort() method calls loseConnection() which tries to shutdown the
TLS connection cleanly. We now call abortConnection() directly which
should promptly close both the TLS connection and the underlying TCP
connection.

I also added some TODO markers to consider cancelling the old previous
timeout rather than checking time.time(). But given how urgently we want
to get this code released I'd rather leave the existing code with the
duplicate timeouts and the time.time() check.

Add some logging for syncing to_device events

Attempt to track down the loss of to_device events
(https://github.com/vector-im/riot-web/issues/2711 etc).

Fix caching on public room list

Update prerequisites instructions on Mac OS

	Fix /backfill returning events it shouldn't

Fix crash in url preview

Signed-off-by: Marcin Bachry <hegel666@gmail.com>

Fix background update that prematurely stopped

Add new API appservice specific public room list

Remove unspecced GET endpoints for e2e keys

GET /keys/claim is a terrible idea, since it isn't idempotent; also it throws
500 errors if you call it without all the right params.

GET /keys/query is arguable, but it's unspecced, so let's get rid of it too to
stop people relying on unspecced APIs.