LU-16830 lod: accurate OSTs iteration for the last speed

LU-16830 lod: accurate OSTs iteration for the last speed

The first fix removed lqr->lqr_start_idx from calculation
of ost_idx for the last speed loop. It was not enough
because lqr->lqr_offset_idx is used at calculation and
also could be changed by another thread. And it leads
to the same OST index during a loop.

1694009994: lod_ost_alloc_rr()) #0 strt 2166 act 0 strp 0 ary 0 idx 0
1694010094: lod_ost_alloc_rr()) #1 strt 2167 act 0 strp 0 ary 0 idx 0
1694010197: lod_ost_alloc_rr()) rc -28

Fixes: cacdaa9251 ("LU-16830 lod: improve rr allocation")
Signed-off-by: Alexander Boyko <alexander.boyko@hpe.com>
Change-Id: Idf4b528dfbf3995cd03580a4214aff9206a52378
Reviewed-on: https://review.whamcloud.com/c/fs/lustre-release/+/52393
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-by: Alexander Zarochentsev <alexander.zarochentsev@hpe.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>