Fix memory tracking for merges and mutations (by destroying earlier)

It fixes only some tiny allocations, and so it should not affect any
huge mutations/merges.

And plus, this should not be a real fix, since peak_memory_usage is
obtained before even destrying this objects, and destroing objects will
unlikely update peak memory usage (although it is possible).

v0: do this in dtors
v2: do this earlier

src/Storages/MergeTree/MergePlainMergeTreeTask.cpp

@@ -87,6 +87,7 @@ void MergePlainMergeTreeTask::prepare()
 
     write_part_log = [this] (const ExecutionStatus & execution_status)
     {
+        merge_task.reset();
         storage.writePartLog(
             PartLogElement::MERGE_PARTS,
             execution_status,

src/Storages/MergeTree/MutatePlainMergeTreeTask.cpp

@@ -40,6 +40,7 @@ void MutatePlainMergeTreeTask::prepare()
 
     write_part_log = [this] (const ExecutionStatus & execution_status)
     {
+        mutate_task.reset();
         storage.writePartLog(
             PartLogElement::MUTATE_PART,
             execution_status,

src/Storages/MergeTree/MutateTask.cpp

@@ -985,6 +985,7 @@ private:
         ctx->mutating_pipeline.reset();
 
         static_pointer_cast<MergedBlockOutputStream>(ctx->out)->writeSuffixAndFinalizePart(ctx->new_data_part, ctx->need_sync);
+        ctx->out.reset();
     }
 
     enum class State