fix: correct distance-to-score conversion for l2 and inner_product metrics in SeekDBStore

[knqiufan]

Summary

• Problem: SeekDBStore.search() hardcoded Math.max(0, 1 - distance) for score conversion, which only works for cosine distance. When configured with l2 or inner_product, similarity scores are incorrect, causing unreliable search ranking and threshold filtering.
• Fix: Store the configured distance metric in the instance and add a distanceToScore() method with metric-specific formulas, matching the Python OceanBase implementation:
  • cosine: max(0, 1 - d/2) — distance range [0, 2]
  • l2: 1 / (1 + d) — distance range [0, +∞)
  • inner_product: clamp((-d + 1) / 2, 0, 1) — negative distance convention
• Add unit tests covering all three metrics plus null/edge cases.

Changed files

File	Change
src/storage/seekdb/seekdb.ts	Add distanceMetric field, distanceToScore() method, update search()
tests/unit/storage/seekdb.test.ts	Add distanceToScore conversion test suite (5 cases)

Test plan

• All 344 existing tests pass (32 files, 7 skipped)
• New distanceToScore tests cover cosine, l2, inner_product, and null input
• Cosine search behavior unchanged (existing score conversion tests still pass)