Elastic Tunnel problems

There are several issues associated with the current Tunnel version:

1. when used together with Squid (OnDemand deployment), it tries to keep idle (unused) tunnel connections for too long, leading to timeouts at tunnel level and errors and tunneled protocol level (SSL handshake errors, wire format negotiation errors, broken pipe errors, class loading errors). Some of these errors cause retries, some take down the agents
2. there are some resource leaks, leading to sockets hanging in CLOSE_WAIT state
3. in some cases, the data forwarding seems to be stuck for a couple of seconds, leading to dropped heartbeats (and probably no bad consequences) - the current assumption is that the tunneled connection needs to be flushed more aggressively.