DAOS-15672 rebuild: Fix pool destroy hangs

See the Jira ticket for the long story. When releasing resources, for
instance, for stopping a pool, we should attempt on all targets, because
the latest version of the pool map may differ from the version used to
acquire the resources. (And in some cases, we no longer have the
original pool map version in the context.)

Signed-off-by: Li Wei <wei.g.li@intel.com>
Required-githooks: true

src/container/srv_target.c

@@ -1823,8 +1823,7 @@ ds_cont_tgt_close(uuid_t pool_uuid, uuid_t hdl_uuid)
 
 	uuid_copy(arg.uuid, hdl_uuid);
 
-	return ds_pool_thread_collective(pool_uuid, PO_COMP_ST_NEW | PO_COMP_ST_DOWN |
-					 PO_COMP_ST_DOWNOUT, cont_close_one_hdl, &arg, 0);
+	return ds_pool_thread_collective(pool_uuid, 0, cont_close_one_hdl, &arg, 0);
 }
 
 struct xstream_cont_query {

src/object/srv_obj_migrate.c

@@ -3162,8 +3162,7 @@ ds_migrate_stop(struct ds_pool *pool, unsigned int version, unsigned int generat
 	arg.version = version;
 	arg.generation = generation;
 
-	rc = ds_pool_thread_collective(pool->sp_uuid, PO_COMP_ST_NEW | PO_COMP_ST_DOWN |
-				       PO_COMP_ST_DOWNOUT, migrate_fini_one_ult, &arg, 0);
+	rc = ds_pool_thread_collective(pool->sp_uuid, 0, migrate_fini_one_ult, &arg, 0);
 	if (rc)
 		D_ERROR(DF_UUID" migrate stop: %d\n", DP_UUID(pool->sp_uuid), rc);