GitLab

A bunch of terrible hacks to test how much speed we would gain at
startup by decrypting events in a batch. That is:
 * Trying to decrypt events once before putting them in the timeline
   to avoid an Event.decrypted for every encrypted event we load from
   disk at startup.
 * When we load batches of events from the store at startup, fetch the
   keys from indexeddb in batches too rather than a separate txn for
   every event (and don't fetch the same session multiple times for
   messages in the same session and the same batch).

Further optimisations that could be applied:
 * Re-use inbound group sessions for multiple messages within the same
   session rather than unpickling a separate one for each message.
 * Not decrypting events until we actually need them
 * Long standing store issues, ie. not storing message data as one
   giant monolithic blob that we have to load at startup.