LU-13614 ldlm: revert LU-11762

LU-13614 ldlm: revert LU-11762

Commit fe5c801657 introduced a problem for recovery.

When recovery timeout reaches hard recovery timeout
target_recovery_overseer() leaves obd_recovery_expired flag set. That
makes check_for_next_transno() to not wait until next replay request
arrives which leads to assertion:
LASSERT(atomic_read(&obd->obd_req_replay_clients) == 0);

Test to illustrace the issue is added.

replay-single.sh:test_59 is added to EXCEPT_ALWAYS list:
  it was broken harmlessly before this patch and this patch made that
  test really fail due to that defect.

Fixes: fe5c80165 ("LU-11762 ldlm: ensure the recovery timer is armed")
HPE-bug-id: LUS-8299
Signed-off-by: Vladimir Saveliev <c17830@cray.com>
Change-Id: Ia694a519b5d73620be3014e92fd671d388550979
Reviewed-on: https://review.whamcloud.com/39532
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>