Inspect: support maintenance bookings and recurring schedules

[JonasScholl]

Changes

app.inspect now supports the full maintenance booking lifecycle: reading and filtering, impact previews, creation, complete updates, locking/unlocking, starting immediately, and deletion. Write operations run through explicit methods, independently of topology transactions.

• Typed specifications, resource targets, schedules, results, and impact reports, with relationships to devices, modules, ports, and edges.
• Maintenance data loads lazily as a snapshot section and reuses data from full snapshots. Successful writes invalidate affected caches without discarding staged topology edits.
• Updates fetch the current revision, preserve the lock state unless explicitly changed, and optionally check expected_rev. Errors include operation and server details; writes are not automatically retried after uncertain transport failures.
• Exact connector allow-list entries for the three maintenance actions, plus detailed anonymized documentation and commented examples.

Recurring schedules

Daily, weekly, and monthly rules expand into dated one-time bookings created in a single request. Each booking has its own server-returned ID and is managed independently. Impact previews evaluate each window separately.

This implementation accounts for behavior observed on VideoIPath 2026.2.0: native recurrence requests produced a broad booking window that could not subsequently be updated. SDK expansion handles IANA timezones and daylight saving time rules, with a limit of 1,000 complete windows per request. Nonempty server-specific iteration filters are explicitly rejected during expansion; transport models preserve these fields. Older server versions have not been verified.

Validation

• poetry run test-unit: 628 tests passed.
• poetry run test-e2e tests/e2e/apps/test_inspect_maintenance.py: 8 live cases passed on 2026.2.0. Coverage includes exact windows, persisted metadata, revision conflicts, lock preservation, start-now, searching and filtering, independent recurring occurrences, module/port/edge targets, open-ended windows, and staged topology edits.
• Live tests use only test-owned mock resources with E2E- labels and the vipat-e2e tag. Cleanup verifies booking removal first, then device removal from Inspect and Inventory; no global sweep is performed.
• New Python files pass Ruff; formatting and Python/JSON examples were checked. The full repository Ruff run still reports existing findings, with no new diagnostics compared with the baseline.
• No credentials or unmodified live captures are included.

Live impact checks use isolated mock resources without services. Populated impact responses are also tested offline.