LU-18351 obdclass: jobstat scaling

LU-18351 obdclass: jobstat scaling

When there is a large number of jobstats (>100k) list walking in
seq_write does not scale and can trigger cpus soft lock-ups.

Prefer red black trees for easier lookup, removal and simplified
tree walk during seq_write()

Add an rbtree with a generated sequential id as jobs are added
to simplify iteration with restarts. The larger the amount of
data being written via seq_write() the more likely and frequent
a restart is needed to seek to new position which can be done
quickly with an rbtree.

Add an lru list to so periodic trim of jobs does not need to
walk the entire tree of jobs.

Unpatched:
  jobs  50000    100000   200000  350000  500000
  time  2.096s   3.672s   7.504s  11.875s 16.721s

Patched:
  jobs  50000    100000   200000  350000  500000
  time  1.366s   3.296s   6.763s  8.895s  11.965s

HPE-bug-id: LUS-12621
Signed-off-by: Shaun Tancheff <shaun.tancheff@hpe.com>
Change-Id: Icca1365af5db761ed89ee9a8a97ce4ded65b8832
Reviewed-on: https://review.whamcloud.com/c/fs/lustre-release/+/56607
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-by: James Simmons <jsimmons@infradead.org>
Reviewed-by: Oleg Drokin <green@whamcloud.com>