Race condition leads to SIGPIPE when sending on a closed socket

[michaelrfraser]

There is currently the possibility of one thread closing a socket, while another thread is trying to call send() on it. Under Linux, this results in a SIGPIPE.

I've tried some basic locking to get around it, but to no avail. Logging an issue for a more detailed look in the future.

See commented TODOs in SocketTest.cpp for more reproduction steps.