Huge refactor to use seperate backend and some other stuff
Catch more exceptions, print more restart logs!
First explicit signature!
Use log everywhere
Do not notify user on rejoin
No one cares about join/part

SMS users rarely do it, and social cues might be in place to handle that
anyway.  Data users flap way to much to be useful, and usually can read
backscroll anyway so they'll see you message when they come back.

If you really want to see join/part messages, you can explicitly ask for
them, with a set debounce (in seconds)
Debound both joins and parts
Advertise jabber:iq:gateway support
Merge presences into old_presences

In case rejoin attempts happen without a sucessful join in the middle.
Rejoin on status 332

Closes #36
Rejoin even if remote server takes awhile to come back

Make sure that on rejoin we end up with a correct list of who is in the

Almost done #36
Replace startup join logic with constant pings

We might restart, the remote might restart, best way is to just
constantly check if we're still in the room.  On startup, re-check each
of our users and rejoin anyone who does not respond to ping.  Then keep

Partial for #36
Persist rooms we create, allow anyone to send invites

The latter is not supported on Prosody
<https://prosody.im/issues/issue/254> which is why we still set everyone
to owner when inviting.
Respond to iq pings
Put time in logs
Room presence manager

We now store on a per-tel basis who they have seen join/part the room.
Joins/parts are stored based on what they have actually seen (via
debounce) and so should always map to SMS we actually sent.

If we get a join from someone we already thought was in the room, that
information is useless.  Do not send it.  Closes #34

We also update the global list in the same place, which I think is now
only used for the re-join on daemon restart.  This is related to #30,
but may not complete it yet.
There can be more than one status

Closes #29
Preserve JID mappings we know about
I think this should prevent unneeded rejoins