[core] Proposed solution for m_ConnectionLock/m_LSLock inversion

[ethouris]

The solution is to prevent applying m_ConnectionLock when the queue is being set the listener.

The same socket will not be attempted to be set listen by another thread because the m_bListening flag prevents this, and it's atomically checked and set under the m_ConnectionLock.

An important logical change is that m_bListening is set true when the listener is ATTEMPTED to be changed, but if this failed, and would be reported by exception, the flag is set back to false. This flag only influences the read-ready flag (which is unimportant for a socket that is not connected) and to remove the listening state when closing. This latter procedure still checks if the listener attempted to be removed is the same as the one that was set and does nothing if not.

Earlier versions: #1824 and #1844