feat(storage): release unused capacity reservations

[levineam]

Storage Janitor callers can now release an unused capacity reservation without claiming any reclaimed bytes. This closes the gap where completed RAID archive runs left active reservations until TTL expiry and could block later work despite consuming no capacity.

This is the public contract slice of SUP-3901. The private macOS runtime will consume the new terminal state in a follow-up integration commit.

Design decisions

• released is terminal across reserve, consume, release, and reap operations.
• Release is idempotent and competes atomically with consume through the existing compare-and-set store boundary.
• Persisted state advances to schema v2. New code upgrades legitimate v1 records in memory, while old v1-only code fails closed on v2 instead of misreading a released reservation.
• The reservation port change is intentionally breaking and documented with the rollback boundary.

Validation

• npm test -- --runInBand in modules/jarvos-storage-janitor: 152 passed, 0 failed.
• Barrier-controlled tests prove both release-wins and consume-wins CAS orderings.
• Managed Opus 5 closure review: no P1/P2 correctness, compatibility, data-integrity, or proof gaps.

Post-Deploy Monitoring & Validation

• Watch Storage Janitor receipts for released, already_released, already_consumed, and store_unavailable dispositions.
• Healthy signal: zero-drawdown runs terminate as released, and later reservations can immediately reuse that headroom.
• Failure signal: active reservations remain after a terminal no-drawdown run, or a v1/v2 schema mismatch is treated as proceed-capable.
• Validation window: the private runtime integration tests before activation, then the separately authorized seven-day mutation-off burn-in.
• Owner: Managed Software Stewardship / Storage Janitor integration lane.

---