Calculation of num_buffers (the max latency in buffers) was up-side-down. If we can allcoate, then our maximum latency match pool maximum number of buffers. Also renamed it to max latency. Finally introduced a min_latency for clarity.