Fixes related to TcpConnector and TcpEndPoint

[pjsaksa]

When TcpEndPoint was closed because of read timeout, TcpConnector::onEndPointClosed() was called repeatedly resulting to mutex deadlock.

Tested with hello_http:

• run hello_http
• open connection with netcat/socat without sending any data
• wait 60 seconds
• hello_http becomes unresponsive

[pjsaksa]

I bundled another bugfix from the same area of code to this PR:

When TcpConnector is destroyed and it deallocates its list of endpoints, those endpoints call TcpConnector::onEndPointClosed() with bad results. This callback should be deactivated during the destruction of TcpConnector.