Bug fix: Use 1 second timeout when closing out connections inside a connection pool lock

[ok2c]

There was a report on the dev list recently about clients getting stuck trying to close out a TLS connection while holding a lock on the connection pool, effectively rendering those clients completely disabled. I thought it should not never happen and it turned out I was wrong. The issue was known and got fixed in core but the fix never got applied to client.

Basically this change-set makes sure the socket timeout is set to 1 second before connections get closed out inside the pool lock.

@arturobernalg Please double-check.

[arturobernalg]

LGTM

[ok2c]

@arturobernalg By the way, if you have spare bandwidth and inclination, there is one potential improvement idea one could experiment with. Presently PoolingHttpClientConnectionManager always closes out connections while holding the global pool lock. Even with a timeout of 1 second that can essentially freeze the clients blocked waiting for a connection lease for over one second, which can be a lot for many applications. What one could do is to queue up connections to be closed with a custom DisposalCallback and then close the queued connections out outside of the pool lock. This approach has its risks but it would be a substantial improvement if it can be made to work properly.