[core] Fix a data race

[yomnes0]

Based on #PR2981
Add a shared mutex on m_pListener to avoid a data race condition.
This fixes #2970

[maxsharabayko]

To keep track, this PR misses the very data race fix originally done in #2961. The proposed fix was in locking m_ConnectionLock before creating a new socket. The blocker was the m_LSLock mutex that was causing lock order inversion.

    try
    {
+       ScopedLock col(ls->core().m_ConnectionLock);
        ns = new CUDTSocket(*ls);
        // No need to check the peer, this is the address from which the request has come.
        ns->m_PeerAddr = peer;

The CSharedObjectPtr and SharedLock are being added in PR #2996. As well as changing the type of the m_pListener to CSharedObjectPtr.