[brainx]

Dashboard observers can read existing durable message receipts through GET /messages, /messages/{message_id} and /messages/capacity, with equivalent /v1 aliases. These routes never dispatch/retry jobs, refresh execution state, recover operations, initialize storage or migrate it.

Stacked on #51; the diff contains only receipt observation, capacity budgeting and their API contracts.

• Bound list pages to 100 rows with an exclusive durable cursor and optional bot filter. Archived receipts remain addressable; prepared receipts are reported as unknown without changing storage.
• Return explicit public-field allowlists, excluding prompts, outputs, endpoints, credentials, fingerprints, upstream keys, leases and internal versions.
• Read existing schema 10 storage in read-only SQLite snapshots. Capacity uses a cooperative two-second SQLite/validation budget and returns a fixed 503 error instead of partial totals. Existing CLI capacity behavior remains unchanged.
• Document complete schemas, pagination/errors, and the Olymp server-side allowlist and pinned-fixture follow-up. No schema migration or message mutation API.

Verification: 85 receipt HTTP/reader/contract, capability, operator-interface, repository-contract and module-size tests passed. A separate 67-test receipt/store/capacity/release/CLI regression run passed. HTTP tests assert unchanged database bytes/SQL contents and prohibit execution/recovery calls. Repository-wide Ruff and formatting passed, mypy passed across 121 source files, Bandit passed, and git diff --check passed.

Limits: capacity's cooperative budget is not a hard filesystem-call deadline; size/free-space observations may be null and can change outside the SQLite snapshot. Normal API access logging and existing API startup initialization remain. No live Olymp deployment or tunnel test was performed.