"unregister_synchronizer" method often breaks when client disconnects

[Gavitro]

When a peer disconnects, it attempts to RPC to other peers to unregister its own synchronizers, but often fails. As a result, the synchronizer maps of the remaining peers don't erase the disconnected peer's synchronizers, leading to memory leaks if left unchecked.

Steps to replicate

• Run at least 2 instances of the demo.
• Host a game on one instance, and connect using the other instance(s).
• Close the client instance(s), leave the host instance running.
• At least one of the instances (I presume the host?) will usually throw the "Attempt to call RPC with unknown peer ID" error and fail to unregister the synchronizers.

[TestSubject06]

I'll need a little time to think this over. It was something I had observed once or twice, but felt it was benign enough to neglect. My rationale was that a handful of potential extra entries in a hashmap wasn't going to be enough of a problem in normal running cases to warrant a rework.

I think it could be relatively quick to fix it by listening for a peer disconnected signal and cleaning up the maps on that. The RPC signals are more intended for in-game comms rather than the entire client exiting.