Shutdown race condition work-around

[SeanMollet]

I doubt this is the best possible solution to the race condition, but I couldn't find anything else that was really viable, since we're waiting for the callback to get called by the main loop.

Adding a small delay (100ms) to TcpTransport::process allows the packets to get processed before we tear down the callbacks, etc.

[SeanMollet]

Fixes #1260

[paullouisageneau]

Thank you but adding a delay is definitely not a solution, especially because this blocks the entire poll service, meaning it blocks reception for all WebSockets.

[SeanMollet]

Better fix provided by Paul-Louis. closing.