[Feature] Point-delete fast path for primary key DELETE to skip the target table scan

[ZZZxDong]

Search before asking

• I searched in the issues and found nothing similar.

Motivation

Currently DELETE FROM on a primary-key table always scans the target table to find matching rows, even when the condition is just primary keys, e.g.:

DELETE FROM t WHERE id IN (1, 2, 3);
DELETE FROM t WHERE id IN (SELECT id FROM keys_to_delete);

For these cases the matched keys are already fully described by the condition itself, so we can build the -D records directly and skip the target table scan entirely. On large tables (we have a case deleting ~10 million keys from a multi-billion-row table by a key table) this is orders of magnitude cheaper than the current scan / join based path.

Solution

Add a fast path in DeleteFromPaimonTableCommand for the pk-upsert delete:

• If the condition is a conjunction of pk = literal / pk IN (literals) covering all primary key columns: extract keys from the condition and write -D rows directly (with a config to cap the driver-side cartesian expansion, falling back to scan when exceeded).
• If the condition is pk IN (subquery) covering all primary key columns: use the subquery result as the key DataFrame, fully distributed.
• Any other condition falls back to the existing scan-based path, so behavior stays unchanged.

Keys absent from the table are harmless since the -D records simply merge away in compaction.

I have a working implementation with tests, will submit a PR.

Anything else?

No response

Are you willing to submit a PR?

• I'm willing to submit a PR!