To reduce lock contention in this mode, makes persistent allocation state per-P,
which means at most 64 kB overhead x $GOMAXPROCS, which should be
completely tolerable.

Change-Id: I34ca95e77d7e67130e30822e5a4aff6772b1a1c5
Reviewed-on: https://go-review.googlesource.com/7740Reviewed-by: Rick Hudson <rlh@golang.org>