Configure TTL on trogon-claims object store bucket

[yordis]

• Store-then-publish in ClaimCheckPublisher can leave orphaned objects if publish fails after a successful put
• The trogon-claims bucket currently uses ..Default::default() with no max_age, so orphans accumulate forever

Design considerations

Two cleanup paths are needed:

Happy path: consumer deletes after resolve

resolve_claim reads the object from the store. After successful resolution, the consumer should delete the object — that's the normal claim check lifecycle (put → publish claim → resolve → delete). This requires adding an ObjectStoreDelete trait and wiring it into resolve_claim or letting the caller handle deletion after consuming the payload.

Failure path: orphaned objects

When publish fails after a successful put, no claim message exists to reference the object. Options:

1. TTL (recommended first step): Set max_age on the bucket (e.g. match stream's 7-day retention). Simple, predictable. Orphans expire automatically. Trade-off: storage cost for the TTL window.
2. Reaper process (if storage pressure warrants): Background job lists objects in the bucket, cross-references against JetStream messages by checking Trogon-Claim-Key headers, deletes objects with no matching claim. More complex but reclaims storage faster.

Recommended approach

1. Set max_age on the trogon-claims bucket config — this is the immediate fix
2. Add ObjectStoreDelete trait and have consumers delete after resolve — this handles the happy path
3. Defer the reaper unless storage pressure becomes a real concern