mon: reduce CPU and memory manager pressure of pg health check

mon: reduce CPU and memory manager pressure of pg health check

When we don't need details about which PGs are stuck, we can just iterate
once over pg_stat map and increase appropriate counters, no need to
repeatedly spam memory manager with stuck_pgs map population (we'll only
use its size). Even if we need details, first check if we actually have any
PGs in any of requested states, so in most cases we won't iterate needlessly
five times over all PG stat map.

This at least halves the time needed by pg health gathering (~3ms,
down from ~6-7ms) on cluster with 31832 PGs and Intel Xeon E5-2640
CPU.

Signed-off-by: Piotr Dałek <piotr.dalek@ts.fujitsu.com>