MINOR: queue: remove the px/srv fields from pendconn

MINOR: queue: remove the px/srv fields from pendconn

Now we directly use p->queue to get to the queue, which is much more
straightforward. The performance on 100 servers and 16 threads
increased from 560k to 574k RPS, or 2.5%.

A lot more simplifications are possible, but the minimum was done at
this point.