Make MatrixClient asynchronous

[non-Jedi]

I haven't tested this yet, and I plan on splitting out the monster commit into
actual bite-size digestible commits, but I'm not entirely sure when I'll get
back working on this again, and there's a chance someone might find it useful.

Also, this is kind of a monstrous PR in scope, so I thought I should let
@Half-Shot know that it was incoming.

I'll sign off on this properly, too once I split out that single commit; don't
worry.

Closes #127

[matrixbot]

Can one of the admins verify this patch?

[non-Jedi]

This PR is substantially ready. It's worked in all manual testing I've done, and
it doesn't break any automated tests. I just need to do the following things
before declaring it not a WIP.

• Write some tests for new asynchronous api (not entirely sure how I'll go
  about this. Any help/advice appreciated)
• Make this into a nice set of commits for review.

The following will be handled in a later PR.

• Make login and registration apis asynchronous.
• Make syncing multi-threaded rather than single-threaded.

Additionally, I have the following question that maybe @Half-Shot or, I dunno,
@erikjohnston (he does python, right? Please ping other members of core team who
may be better suited to answer questions for this sdk) might be able to help with:

1. Should I write a series of locks for anything that is get/set from the
   matrix homeserver or should that be handled at a higher level? (e.g. if
   there's an ongoing request to set displayname, should that stop any
   requests to get displayname until it finishes?) EDIT: After talking to
   @richvdh:sw1v.org in #matrix-dev:matrix.org, I will not be implementing
   these sorts of locks--at least not yet.

[4nd3r]

could this PR pave road to automatic (in SDK) connection retrying to HS after connection loss? or #123?

[non-Jedi]

@4nd3r this could definitely make that easier. #123 was a typo that I've now
fixed to #127. Most of the issues people have with retrying are with the
sync thread, which is actually already supposed to retry, but thanks to the
bug in #131, doesn't really.

But if what you're looking for is automatic retry for all requests, yes,
making the client async means we just have to implement that once within the
queue class.

[non-Jedi]

Note that this will make fixing #6 significantly easier.

[subins2000]

Any updates on this ?

[non-Jedi]

@subins2000 so kind of the state of the project right now:

I'm waiting on @Half-Shot (or anyone else familiar with the layout of the project) to review #160. After that, I'm planning on doing a 0.1.0 release, and then after the release, I'll be polishing this up (getting rid of merge conflicts), adding some tests, and then getting this reviewed to potentially merge.

[non-Jedi]

I don't think this is the right approach any more. Would prefer to build async on top of #168 once that's merged and make it exclusive to 3.5+.