kernel - Fix numerous performance problems with the pageout daemon
* The VM page queues were not being fully utilized, causing the pageout
daemon to calculate incorrect average page counts for deactivation/freeing.
This caused the pageout daemon to dig into the active queue even when it
did not need to.
* The pageout daemon was incorrectly calculating the maxscan value for each
queue. It was using the aggregate count (across all 256 queues) instead of
the per-queue count, resulting in long stalls when memory is low.
* Clean up the PQ_L2* knobs, constants, and other cruft, reducing them to
the essentials for our goals.
Reported-by: vsrinivas, thesjg, luxh, etc