Workflow [PR], commit [14076d5]

The failed stateless tests are also related to enable_producing_buckets_out_of_order_in_aggregation. The old behaviour without that setting is not the same as with this setting enabled. To prove that, we can see the diff of the PR that introduced that setting: https://github.com/ClickHouse/ClickHouse/pull/80179/changes#diff-1806a5c1f13b491c615e887366acf1daac6d85b1528be423e3762705482c80fbR574

Otherwise the actual output of the test shows that the issue has been resolved, which is seen by the line GroupingAggregatedTransform 60 → 1 for ReadMergeTree which matches the upstream expected output and proves the pipeline is not being narrowed.

I'll update the .reference file to let the test pass, it's a matter of adding a new pass to it.

QA verification

This change fixes a logical error in distributed two-level aggregation when using Merge over Distributed tables.
The fix and the new stateless test were reviewed and approved upstream in clickhouse/clickhouse.

For the Altinity backport:

• All upstream CI checks passed, including the new stateless test 03403_distributed_merge_two_level_aggregation
• Altinity regression tests passed

Considering that the stateless test was adapted for Antalya 25.8 (without enable_producing_buckets_out_of_order_in_aggregation),
I also executed the test 03403_distributed_merge_two_level_aggregation locally to confirm the expected behavior.

The test was executed:

• on the PR build (antalya-25.8.14.20001)
• and on the latest Antalya release (25.8.12.20747)

This confirmed the intended behavior:

• before the fix, the pipeline was narrowed (GroupingAggregatedTransform 8 → 1)
• after the fix, the pipeline is no longer narrowed (GroupingAggregatedTransform 60 → 1), matching the expected behavior.

The change is well-scoped, guarded by specific conditions, and covered by automation.

Marking this PR as verified.