fix issue when port is left dangling and catch poolboy timeout errors

[criticalbh]

When we open connection to Faktory, we observe some events but we never take care of port in terms of closing it.

When we receive one of following signals : :closed, :enotconn, :econnrefused we just ignore current socket and open a new one, which leads to port leak.
In this PR we are explicitly closing socket, and after observation :erlang.ports it shows that number of ports are not rising any more, everything stays in normal boundaries as it should be.

Also, we have handle_push_result({:error, :timeout}, job) but we are not catching timeout properly, because poolboy checkout raises an exception. In this PR we are catching exception and returning {:error, :timeout} which should be picked up by handle_push_result.