fix(retrieval): trace the dense channel threshold for hybrid too

[mrsibe]

Stacked on #199 (feat/eval-sweep-dashboard). Retarget to main after the chain merges.

What does this PR do?

Fixes a trace that misdescribed what retrieval actually did. Found in the #192 review.

The bug

hybrid runs a dense leg, and that leg applies a similarity floor. The trace recorded threshold: undefined, while DenseRetriever.candidateHits() was quietly applying request.threshold ?? 0.5:

actual behaviour:   hybrid ├─ dense  threshold = 0.5
                           └─ BM25   no threshold
trace recorded:     strategy = hybrid, threshold = undefined

So the #157 promise — that a retrieval snapshot reproduces the retrieval it describes — does not hold for hybrid. Harmless while dense is the shipped strategy; a real problem the moment hybrid becomes one.

The cause and the fix

Two separate ?? 0.5 defaults: one inside candidateHits(), one implicit in what the trace wrote. There is now one function, denseChannelThreshold(strategy, requested), and the same value is both handed to the dense channel and written to the trace, so the two cannot drift apart again:

• dense / hybrid → the requested threshold, or DEFAULT_DENSE_THRESHOLD
• sparse → undefined (BM25 has no similarity floor)

The trace field is renamed threshold → denseThreshold. A bare threshold on a strategy: 'hybrid' trace reads as "the threshold for all of hybrid", and that ambiguity is exactly what hid the bug. parseRetrievalSnapshot reads the legacy threshold from already-persisted snapshots as a dense threshold — the value was always the dense leg's, so the backfill states what that retrieval actually did rather than inventing a default.

Testing

• npm run typecheck — clean
• npm test — 502 pass, with new coverage for all three cases above plus a regression test asserting hybrid traces 0.5 while sparse traces nothing

Not covered

HybridRetriever needs a vector store, so it has no unit test here. The pure decision is pinned instead, and the wiring uses a single variable — which makes the divergence impossible rather than merely tested against.

Related

Part of #192 (review follow-up).