evaluate_claim should return full per-claim eval intent (mode + params)

[aural-psynapse]

Why

evaluate_claim currently returns the verification_mode and the result, but not the per-mode parameters Cluster A chose (json_path, expected_json_schema, range_min, range_max, query_record_id).

That makes downstream audit/replay incomplete: a consumer storing the verdict can record that a claim CAUGHT but not how the comparison was framed. Without those fields, replaying a failed run requires keeping the original HandoffClaim somewhere external to the verdict.

What

Extend the dict returned by provably.handoff.eval_modes.evaluate_claim (and the corresponding return shape used in handoff/_http.py) to include:

• json_path
• expected_json_schema (nullable)
• range_min, range_max (nullable)
• query_record_id

verification_mode is already there. Existing result, claimed, indexed, indexed_at_path, detail keys stay as-is.

Acceptance

• New keys present on every returned verdict (null when not applicable).
• Tests in tests/unit/test_eval_modes* updated to assert the new keys.
• No change to HandoffClaim shape; this is purely the eval response.

Downstream (out of scope here)

Once this lands, the monorepo can extend its agent_steps audit table to persist these fields. Tracked separately if needed.