Add whitelisting of users before transport starts

[andrevmatos]

UDP is noop for now, but for matrix, ensure invites sent while offline are properly handled
Fix #2779 , as per 2. of #2779 (comment)

[LefterisJP]

@andrevmatos Don't say that this fixes #2779 since we can't confirm this without a reproducible test case. The way it's worded now it will close #2779 if merged.

[ulope]

I tested this with one of the hanging scenarios from before (with an existing token and node state) and it immediately still hung on the first transfer.

Edit: I haven’t checked the node logs though. It’s still possible that this fixes the communication issue but we could for example run into the nonce mismatch problem.

Will check when I’m home.

[hackaugusto]

AFAICR this is not fixing the problems we discussed.

We must call _initialize_message_queues before the transport handles incoming messages. This is necessary to ensure the messages stay in the correct order, otherwise there is race condition where incoming messages can trigger new Send events which end up scheduled in before the ones in the queue.

With that said, this really needs proper documentation. If I recall correctly we are doing the whitelisting prior to starting the transport because otherwise valid invites would be lost, there is not mention in the code or in the commits about the race and how it's being fixed.

[andrevmatos]

Included and superseded by #2948 . @hackaugusto 's comments will be addressed there.