Problem

There is no mechanism to identify or remove unreachable objects. Storage grows monotonically — deleted branches, force-pushes, and amended commits leave orphaned objects forever.

Impact

• Storage costs grow without bound
• No way to reclaim space after branch deletion or history rewrite
• S3 backend particularly affected due to per-object storage costs

Acceptance Criteria

• Reachability walk from all refs to identify live objects
• Pruning of unreachable loose objects older than a configurable grace period
• Works across all three storage backends
• Dry-run mode that reports what would be deleted
• Safety: never deletes objects reachable from any ref
• Tests covering: orphan cleanup, grace period, concurrent write safety