Fix test_system_merges by using mutations_sync=1

After early_constant_folding started to ignore not only ignore(), but
all functions with isSuitableForConstantFolding() == false, there became
more sleep(2) calls for this test:
- MergeTreeDataSelectExecutor::readFromParts -> DB::KeyCondition::KeyCondition
- MergeTreeDataMergerMutator::mutatePartToTemporaryPart -> DB::isStorageTouchedByMutations -> FilterTransform::transform
- MergeTreeDataMergerMutator::mutatePartToTemporaryPart -> DB::MergeTreeDataMergerMutator::mutateAllPartColumns -> FilterTransform::transform

While before it was optimized to 0 during WHERE analysis.

tests/integration/test_system_merges/test.py

@@ -134,7 +134,9 @@ def test_mutation_simple(started_cluster, replicated):
         result_part = "all_{}_{}_0_{}".format(starting_block, starting_block, starting_block + 1)
 
         def alter():
-            node1.query("ALTER TABLE {name} UPDATE a = 42 WHERE sleep(2) OR 1".format(name=name))
+            node1.query("ALTER TABLE {name} UPDATE a = 42 WHERE sleep(2) OR 1".format(name=name), settings={
+                'mutations_sync': 1,
+            })
 
         t = threading.Thread(target=alter)
         t.start()
@@ -159,8 +161,6 @@ def test_mutation_simple(started_cluster, replicated):
         ]
         t.join()
 
-        time.sleep(1.5)
-
         assert node_check.query("SELECT * FROM system.merges WHERE table = '{name}'".format(name=table_name)) == ""
 
     finally: