Handle coalesce partition handle when bucket count does not match but compatible

[rongrong]

Should we have release note for this? It's technically a fix. Also do we have release note for default session value changes?

== RELEASE NOTES ==
General Changes
* Fix query failures with incompatible partition handle when session property ``partial_merge_pushdown_strategy`` is set to ``PUSH_THROUGH_LOW_MEMORY_OPERATORS`` and ``optimize_full_outer_join_with_coalesce`` is set to ``true`` and query has mismatched partition and uses ``FULL OUTER JOIN`` with ``COALESCE``.

[rongrong]

cc @caithagoras

[caithagoras]

Should we also update the error message?

[caithagoras]

Consider squashing the commits into one: You can add descriptions about default value changes into the commit message of the 3rd commit.

[caithagoras]

We should have release notes for the fix. Something like

General Changes
* Fix query failure for ``CROSS JOIN`` with ``COALESCE`` when bucket counts are
  compatible but mismatched.

[caithagoras]

Thanks for the fix! lgtm, will let @wenleix take a look.

[wenleix]

LGTM % minor comment.

[wenleix]

what about:

    this.handle.equals(other.handle) || metadata.isRefinedPartitioningOver(session, this, other) || metadata.isRefinedPartitioningOver(session, other, this)

[wenleix]

what about

   && metadata.isRefinedPartitioningOver(session, probeProperties.getNodePartitioning().get().getHandle(), buildProperties.getNodePartitioning().get().getHandle()).isPresent()) || metadata.isRefinedPartitioningOver()...

You might want to save probeProperties.getNodePartitioning().get().getHandle() and buildProperties.getNodePartitioning().get().getHandle()).isPresent() as local variable.

[wenleix]

optimize_mismatched_bucket_count cannot be default to true, this is because engine cannot reason whether the "compatible bucket" will result in more memory usage or not, for example:

          AGGR
            |
          JOIN
         /     \
 TS(2048)    TS(128)

Now it will be executed as if there are 128 buckets, and cause increase in local memory usage. This is the motivation to deprecate optimize_mismatched_bucket_count and introduce refined partitioning in #12611, so engine can reason why the partitioning is refined or coarsened .

[wenleix]

LGTM.