obs: scheduled-lane health, freshness telemetry, and standard artifact envelope

[thymikee]

Part of #1412 (Track E — gate integrity). From external review: nightly/weekly lanes (replays, perf, conformance differential, and the upcoming fuzz/mutation/torture/model-bench lanes) can fail or stop running for weeks while PR CI stays green. The observatory must watch the watchers.

Deliverables

• Standard artifact envelope for every scheduled lane: schemaVersion, commit SHA, tool/config versions (or content hashes), seed where applicable, duration, result. Existing nightlies adopt it as they are touched; new lanes (test: nightly fuzz lane — parser inputs must fail as typed AppErrors, never hang #1414, test: weekly mutation-score ratchet over decision kernels (Stryker, non-gating first) #1415, test: seeded concurrency torture lane for session/lease/lock invariants #1416, ci: weekly scheduled model-backed help-bench run with pass-rate deltas #1427) MUST ship with it.
• A scheduled health job: reads last-run/last-success for every schedule:-triggered workflow via the GitHub API (the workflow list is derived from .github/workflows/, not hand-maintained) and posts/pings a tracking issue when any lane misses or fails two consecutive cadences.
• Retry telemetry feeding the repo-health snapshot (obs: repo-health snapshot command aggregating existing analyzers into one JSON #1423) and delta comment (obs: evolve the size-report PR comment into a quiet quality-delta comment + metric history on main #1424): retry count per run and retry-eligible-file count (ci: single-retry policy for enumerated contention-flaky files (timeouts only) #1419) as ratcheted metrics — the flake list shrinking is the goal, and growth must be visible.
• Timeout classification from structured Vitest reporter metadata, never error-text sniffing (the classifier lessons: categories from recorded fields, not parsed names).
• Branch-protection required-contexts audit lives here if it needs elevated API scope (see the gate-manifest sibling issue).

Acceptance

• Simulate a lane going dark (disable a test schedule twice): the health job opens/pings the alert issue.
• Every scheduled artifact produced after this lands carries the envelope; obs: repo-health snapshot command aggregating existing analyzers into one JSON #1423's snapshot exposes lane freshness fields.

Consolidation required (PR-wave review, 2026-08-01)

Three PARALLEL implementations of this issue's health watcher shipped as riders in the first worker wave — #1438 (scripts/scheduled-lane/, 904 LOC), #1439 (scripts/scheduled-lane-health/, 1,064 LOC), #1441 (scripts/lane-health/, 472 LOC) — two of them creating the SAME workflow filename (.github/workflows/scheduled-lane-health.yml) and two granting issues: write without dedicated review. All three PRs have been asked to strip their watcher.

The dedicated PR for this issue must therefore START by comparing the three implementations and consolidating one (steal the best pieces; the others die). Additional requirements from the review:

• The shared envelope module home is scripts/lib/lane-envelope.ts (from test: ratchet mutation score over enumerated decision kernels #1441, the cleanest of the four envelope writers in the wave); test: nightly fuzz lane — parser inputs must fail as typed AppErrors, never hang #1414/test: seeded concurrency torture lane for session/lease/lock invariants #1416's lanes rebase onto it.
• The issues: write grant gets its own explicit review attention in this PR — it is the reason riders were unacceptable.
• The watcher's workflow enumeration must stay derived from .github/workflows/ (all three implementations should be checked for hand-maintained lists before adoption).