feat(incidents): the timeline leads with what is worst, not what is newest

[passcod]

🤖 The incident timeline was ordered by time alone, so an operator opening a page mid-incident had to read a chronology to find the failure — and a warning filed a minute ago sat above a failure that had been going for an hour.

Issues are now ordered by the severity of their effective result, ties broken by most recent, with notes below every issue:

✖ failed   disk_free      2h ago
✖ failed   db_connect     6h ago
▲ warning  pg_tuning     10m ago
● passed   sync_lookup    1h ago
─────────────────────────────
🗒 note  felix            5m ago

Effective rather than observed, so a check an operator capped at warning ranks as a warning here too — the same result the rest of the incident workflow acts on. The rank comes from CHECK_RESULT_ORDER, already the UI's severity vocabulary, rather than a second ordering to keep in sync.

An issue with no recorded result predates the check-state model and has no severity to rank on, so it sorts below everything graded — where indexing an unknown directly would have put it first.

Coverage

Two Playwright specs, seeded so that time-ordering alone would produce the opposite order: one over the full timeline (failure above warning above recovered above note), one over the Issues filter. Assertions key off the issue message, which is the row's always-visible summary — the check name only renders in the expanded provenance line, and recovered issues render collapsed.

seedIncidentNote is new; notes had no seed helper.