Don't close connection with query timeout when no query is in flight

[jcoleman]

Fixes #579.

[petere]

I'm not sure this is the right solution. It seems unprincipled to mix in a condition based on the server's transaction status. I think the problem here is rather that the query timeout is calculated based on the client's request_time rather than the query_start, and moreover it should check whether query_start != 0, that is, is a query in progress at all. At the moment, it is more an activity timeout rather than a query timeout.

Note that there is a second piece of code that handles query_timeout, in pool_client_maint(). There, it uses a mix of request_time and query_start. This is also confusing.

[JelteF]

I agree with Peter here that this is currently not the right fix, also this should have a test.

[JelteF]

To clarify, I agree that there is a bug here. But the we should not fix it in the way that this PR currently does. The suggestion from @petere seems like a better approach.