Branch b1_6

Add a new item(lq_lock) in struct lustre_qunit which protects every single qunit
and qunit_hash_lock only protects qunit hash. It will improve parellelism.
i=johann
i=andrew