parallel_ces produces N× inflated results for EXISTS/NOT EXISTS with self-join tables

When parallel_ces RBO rule is applied and a query contains EXISTS/NOT EXISTS with a self-join (same physical table, different aliases), results are inflated by exactly parallel_factor×.

Repro:

SET @@columnstore_unstable_optimizer=ON;

SET @@columnstore_query_accel_parallel_factor=5;

SELECT COUNT(*) FROM lineitem l1

WHERE EXISTS (SELECT * FROM lineitem l2 

  WHERE l2.l_orderkey = l1.l_orderkey AND l2.l_suppkey <> l1.l_suppkey)

AND l1.l_receiptdate > l1.l_commitdate;

-- Expected: 366216, Actual: 1831080 (5×)

• factor=1 → correct, factor=N → N× inflation
• Only ExistsFilter affected (IN/SelectFilter works correctly)
• Only self-joins reproduce (EXISTS on different table is fine)
• Impacts TPC-H Q21 (numwait = parallel_factor × correct value)

Regression: Correct in b5e11f2be (2025-11-06). Wrong values baked into .result by 4d13bd51b (2025-12-09).

Possible root cause: ColumnStatistics stores stats by {{

}} without alias. For self-joins both aliases share one entry with the alias of whichever table was registered first. MariaDB processes subquery tables first, so l2 registers before l1. makeUnionFromTable() gets a keyColumn with alias="l2" and creates range-partition filters on l2.l_orderkey inside UNION ALL units that only scan l1 → filter is a no-op → all rows in every partition → N× duplication.

Fix: In makeUnionFromTable() (rbo_apply_parallel_ces.cpp), align keyColumn alias with the target table after obtaining it from statistics.