We want to wait until we have read the response body before we log the request
as complete, otherwise a confusing thing happens where the request appears to
have completed, but we later fail it.

To do this, we factor the salient details of a request out to a separate
object, which can then keep track of the txn_id, so that it can be logged.

Use the actual origin for push transactions, rather than whatever the remote
server claimed.

This manifested in synapse repeatedly setting the tag for the room

This was due to a bug where we mutated a cached event's contents

Splits the state_group_cache in two.

One half contains normal state events; the other contains member events.

The idea is that the lazyloading common case of: "I want a subset of member events plus all of the other state" can be accomplished efficiently by splitting the cache into two, and asking for "all events" from the non-members cache, and "just these keys" from the members cache.  This means we can avoid having to make DictionaryCache aware of these sort of complicated queries, whilst letting LL requests benefit from the caching.

Previously we were unable to sensibly use the caching and had to pull all state from the DB irrespective of the filtering, which made things slow.  Hopefully fixes https://github.com/matrix-org/synapse/issues/3720.