osd: Update PGSnapTrim op queue item cost to reflect average object size

osd: Update PGSnapTrim op queue item cost to reflect average object size

Previously, a static value of snap_trim_cost (1 MiB by default) for
PGSnapTrim item was used (see config option osd_snap_trim_cost). For pools
with significantly different sizes of objects, the static cost doesn't
accurately estimate the amount of IO each snap trim operation requires.
Instead, add a cost_per_object parameter to OSDService::queue_for_snap_trim
and set it to the average object size in the PG being queued by using
PG::get_average_object_size().

In addition, for the mClock scheduler, the cost_per_object is multiplied
by the actual number of object trimmed per iteration. This multiplier is
represented by osd_pg_max_concurrent_snap_trims config option which is
used when the actual work starts (See DoSnapWork).

Note: The above cost calculation is valid for most snap trim
iterations except for:

1. The penultimate iteration which may return only 1 object to be trimmed,
in which case the cost will be off by a factor equivalent to the average
object size.

2. The final iteration (returns -ENOENT), involving clean-ups.

Fixes: https://tracker.ceph.com/issues/63604
Signed-off-by: Sridhar Seshasayee <sseshasa@redhat.com>
(cherry picked from commit fbd5c40edccccbb44b66e9d82fc71b14ac1d04ae)