Fix Matins hymn to actually keep the Assumption's octave hymn all week

resolveMatinsHymn's own doc comment already claimed the octave "keeps
the feast's own proper hymn all week," but the code never implemented
it -- it only matched when the day's own winner literally was
`assumption` (Aug 15 itself). Every other octave day (16, 17, 19, 20,
21, none of whom has a Matins hymn of their own authored) fell straight
through to the plain ferial hymn instead.

Adds a real octave-fallback tier between the existing per-feast override
and the seasonal tier. A deliberate departure from the reference engine,
which doesn't do this either (Monastic 1617's octave days have no
[Hymnus Matutinum] override at all) -- consistent with how this project
already treats octave readings/commemorations more generously than any
one source track.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VZSAgRi4QE4XRTqVto93zA

TODO.md

@@ -2296,6 +2296,38 @@ two years' Easter dates, and `tests/calendar/august-sanctoral.test.ts`'s existin
 was updated to match (the day now reverts to the plain Assumption octave, same shape as St.
 Lawrence's own Aug 12 case). `npm test` (429 tests) and `tsc --noEmit` both pass.
 
+### Assumption-octave Matins hymn fixed (2026-08-22)
+
+Separately, the "no antiphons and a weird hymn today" complaint that prompted the investigation
+above turned out to be a distinct, narrower bug: `resolveMatinsHymn`'s own doc comment
+already claimed the Assumption's octave "keeps the feast's own proper hymn all week," but the
+code never implemented that — it only fired when the day's own winner literally *was*
+`assumption` (Aug 15 only). Every other octave day (16, 17, 19, 20, 21 — St. Joachim, St.
+Hyacinth, St. John Eudes, St. Bernard, St. Jane Frances de Chantal, none of whom has a Matins
+hymn of their own authored) fell straight to the plain ferial hymn. Fixed by adding a real
+octave-fallback tier (`resolveActiveOctave(day.date)` → `matins-hymn-${octave.id}`) between the
+existing per-feast override and the seasonal tier. Confirmed against the reference engine that
+Monastic 1617 itself doesn't do this either (no octave day there has its own `[Hymnus
+Matutinum]` override) — this is a deliberate departure, not a restoration, consistent with how
+this project already treats octave readings/commemorations more generously than any one source
+track. Aug 18 (no sanctoral winner) and Aug 22 (once IHM moved) needed no change — both already
+reached `matins-hymn-assumption` via the pre-existing `resolveOfficeWinner` octave-substitution
+path, which only fires when the day's own winner isn't sanctoral.
+
+**Not fixed, logged instead** (user decision): the deeper reason Matins had *zero* nocturn
+antiphons at all on Aug 22 — no `matins-psalmody-overrides` entry exists for `assumption` or
+`immaculate-heart-of-mary` (or, generally, any Duplex-2-classis+ feast lacking one), so Matins
+falls all the way to the bare ferial psalm table, which by design carries no antiphons — is a
+broader content gap than today's date, tracked in the "Matins psalmody overrides" bullet above
+rather than authored now. Also newly true because of this change: `hours/matins.ts`'s psalmody
+override lookup is hard-gated to `winner.kind === 'sanctoral'`, so even if IHM's psalmody were
+authored later, it wouldn't apply as a temporal-kind winner without that gate being loosened
+too — noted for whoever picks this up.
+
+New tests in `tests/hours/matins.test.ts` cover the fallback across Aug 16-22, including the
+Aug 17 edge case (correctly still ferial, since that day is actually governed by St. Lawrence's
+own octave, not the Assumption's). `npm test` (430 tests) and `tsc --noEmit` both pass.
+
 ## Known, deliberate simplifications (not bugs — working as designed)
 
 - `getDayCollects`: each collect in a multi-collect day renders as its own