LU-16571 utils: fix parallel "lfs migrate -b" on hard links

LU-16571 utils: fix parallel "lfs migrate -b" on hard links

Multiple blocking "lfs migrate" on the same file can exhaust "ost"
service threads of an OSS CPT.

llapi_get_data_version(...,LL_DV_RD_FLUSH) causes the OSS server to
take a server-side extent lock PR to force clients with write lock to
update the data version of the object.

migrate_block() (lfs.c) checks the file data version is check with
LL_DV_RD_FLUSH before taking the group lock.
So "ofd_getattr_hdl()" server side lock will conflict with the lfs
instance that has the group lock.
Each attempt to get server-side extent lock will take an "ost" service
thread slot waiting the group lock to be released.

If all threads of the "ost" servive are exhausted on a CPT, the OSS
can not handle requests from the client and it will get queued inside
the NRS policy. This causes the lfs process with the group lock to
hang (pread needs "ost" service to get sizes of objects).

This patch check the file data version inside the group lock without
LL_DV_RD_FLUSH. This flag is not needed, the client already has an
extent group lock on all the OST objects.

Add the regression test sanity 56xj.

Test-Parameters: testlist=sanity env=ONLY=56xj,ONLY_REPEAT=20
Test-Parameters: testlist=sanity env=ONLY=56
Test-Parameters: testlist=sanity env=ONLY=56
Test-Parameters: testlist=sanity env=ONLY=56
Signed-off-by: Etienne AUJAMES <eaujames@ddn.com>
Change-Id: I0bacd372dd6f36a4ac776133dff45dc836c7c7f7
Reviewed-on: https://review.whamcloud.com/c/fs/lustre-release/+/50113
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-by: Patrick Farrell <pfarrell@whamcloud.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>