fix(sql-parser): preserve SQL top-k secondary ordering semantics #682

Parent

#678

What to build

Prevent SQL heap top-k detection from changing the meaning of queries with additional ORDER BY terms. The current detector validates only the primary aggregate ordering, while heap top-k execution bypasses normal SQL post-processing.

Either reject unsupported secondary ordering during top-k detection or implement the complete ordering behavior, including ties.

Acceptance criteria

• Queries with unsupported secondary ORDER BY terms are rejected or handled correctly.
• Aggregate descending ordering remains supported.
• Tie ordering is deterministic and matches the documented SQL behavior.
• Tests cover secondary ordering, ties, and LIMIT interaction.

Blocked by

None - can start immediately.

Concrete example

This query has a meaningful secondary ordering:

SELECT srcip, SUM(bytes) AS total
FROM netflow
GROUP BY srcip
ORDER BY total DESC, srcip ASC
LIMIT 2;

The detector currently accepts it because the first ordering item is total DESC. The execution path then skips normal SQL post-processing, so ties are ordered by the heap/result implementation rather than by the requested srcip ASC.

A minimal fix could reject any top-k query where order_by.len() > 1; a broader fix would carry all ordering terms through execution.