Implement bounded resolved cleanup batching optimization

[jununfly]

This was generated by AI during triage.

Parent

Plan: docs/plans/2026-06-20-rust-hybrid-finalization-cleanup-diagnostics-and-batching.md

What to build

Attempt one bounded TypeScript-side optimization of the resolved-reference cleanup path. The optimization may change batching or SQL shape for deleting unresolved-reference rows that were successfully resolved, but it must not change which references are deleted.

Do not introduce Rust core execution, do not change schema, do not change edge insert ownership, and do not touch reference disambiguation.

Acceptance criteria

• Only the resolved-reference cleanup path is optimized.
• Intentionally unresolved cleanup behavior remains unchanged.
• Cleanup contract tests from Add deterministic cleanup contract tests #327 pass.
• Graph output is stable on the deterministic fixture.
• Fallback taxonomy is stable on the deterministic fixture or any targeted smoke used for this issue.
• No SQLite schema changes.
• No changes to disambiguation, import/framework resolution, dynamic-dispatch synthesis, confidence, ranking, or resolvedBy.
• If the optimization is ineffective or unsafe, record a no-go note instead of broadening scope.

Blocked by

• Split finalization write and cleanup profile diagnostics #326
• Add deterministic cleanup contract tests #327

[jununfly]

This was generated by AI during triage.

Implemented the bounded attempt. Resolved cleanup now uses a compact rowid-range deletion path while intentionally unresolved cleanup remains on the previous rowid deletion path. Deterministic tests pass and the closeout records this as no-go as a standalone performance lever rather than broadening scope.

[jununfly]

This was generated by AI during triage.

Closing as completed with no-go conclusion for the bounded optimization as a standalone performance lever. The implementation remains narrow and does not broaden into edge write or Rust subprocess cleanup.