fix(of-lectionary): stop serving drifting readings for O-Antiphon and Christmas-season dates

The bootstrap excluded 19 lectio bases as "date-keyed duplicates", but
17 of them were not duplicates at all: the 8 O-Antiphon days
(17-24 December) and 8 further Christmas-season dates (29-31 December,
2-5 and 7 January) each carry unique per-date content found nowhere else
among the emitted entries, verified against lectio's own ini directly.
17 December is Gen 49:2,8-10 / Matthew 1:1-17, but colitur was falling
through to that year's unrelated weekday-keyed "advent-3-mon" content
instead.

OLM n. 69.3 explains why: those ferias are fixed by civil date, unlike
every other Advent/Christmastide/Paschaltide feria, which is merely
non-alternating within a weekday slot. Temporal_of's own ferial slugs
for these dates are weekday-keyed, so looking one of those up served
whichever OTHER date happened to share that year's weekday alignment --
a citation that silently drifted year to year on ~16 real civil dates
(6 January stays excluded: Temporal_of fixes Epiphany there
unconditionally, so that date can never reach the ferial path at all;
two more, easter-6-thu and advent-4-sat, stay excluded for their own,
different structural reasons).

Fixed with a date-keyed lookup route (Lectionary_of.date_keyed_slug),
tried before the weekday-keyed one in readings' own step 3.
Temporal_of's slugs are unchanged -- only which lectionary key resolves
a day's citations. tools/bootstrap_lectionary_of.ml now maps the 17
bases to synthetic date-keyed slugs instead of excluding them, and its
own reachability sweep calls date_keyed_slug directly so the two can
never drift apart. Re-emitted data/of/lectionary.sexp (754 -> 770
entries) and corrected its provenance header, which previously
described these dates as excluded duplicates.

Also states, for the first time, a Minor finding from the same review:
190 of 222 shipped sanctoral slugs have no dedicated lectionary entry
and correctly fall through to the ferial per OLM norms -- expected
behaviour, not a gap, but never said in the coverage report before.

New tests pin 17 December (Gen 49:2,8-10 / Matthew 1:1-17) and 2 January
(1 John 2:22-28 / John 1:19-28) in two years with different weekday
alignments, proving neither reading drifts, plus direct boundary
coverage of date_keyed_slug itself (both windows, the 6 January
exclusion, and the Sunday guard).

1 files changed, 48 insertions, 2 deletions

diff --git a/lib/rites/rite_of/lectionary_of.mli b/lib/rites/rite_of/lectionary_of.mli
index 0c4d1d6..4a447d2 100644
--- a/lib/rites/rite_of/lectionary_of.mli
+++ b/lib/rites/rite_of/lectionary_of.mli
@@ -110,8 +110,13 @@ val weekday_cycle : year_start:(int -> Date.t) -> Date.t -> weekday_cycle
       closed kernel type outside this task's own scope to extend; see
       lectionary_of.ml's own implementation comment on this branch for the
       full argument. [said] the observed slug.
-    - {b Step 3} -- the day's own temporal slug, same three-way lookup.
-      {!Colitur_kernel.Mass_formulary.Own_slug}, [said] that slug.
+    - {b Step 3} -- the day's own temporal slug. FIRST tries
+      {!date_keyed_slug} (OLM n. 69.3's date-fixed windows -- the
+      O-Antiphons and the pre-Epiphany run of the Christmas season), THEN
+      falls back to the same three-way slug lookup step 2 uses.
+      {!Colitur_kernel.Mass_formulary.Own_slug} either way, [said] the
+      temporal office's own slug -- {!date_keyed_slug} changes which KEY
+      resolves the citations, not which office is said to have said them.
     - A day matching neither gets [(None, [])] -- see data/of/lectionary
       .sexp's own provenance header for exactly which days that is, both
       directions, measured, not guessed.
@@ -134,3 +139,44 @@ val readings :
   date:Date.t ->
   temporal_at:(Date.t -> (Vocab_of.season, Vocab_of.rank) Temporal.t) ->
   Mass_formulary.t option * Citation.t list
+
+(** OLM 1981 Praenotanda n. 69.3 (the same passage {!weekday_cycle}'s own
+    doc comment cites: "Pro feriis Adventus, temporis Nativitatis et
+    paschalis cyclus eodem modo annualis est: lectiones ideo non
+    mutantur" -- Advent/Christmastide/Paschaltide ferial readings do not
+    change year to year). Every OTHER such family is non-alternating
+    WITHIN a weekday slot ({!weekday_cycle}); two short windows go
+    further and are keyed by CIVIL DATE instead: the O-Antiphon days
+    (17-24 December) and the pre-Epiphany run of the Christmas season
+    (29-31 December, then 2-5 and 7 January) -- verified concretely
+    against lectio's own ini, not merely inferred from the citation: 17
+    December is Gen 49:2,8-10 / Ps 72:1-2,3-4ab,7-8,17 / Matthew 1:1-17
+    regardless which weekday it falls on in a given year.
+    {!Rite_of.Temporal_of}'s own ferial slugs for these same civil dates
+    are WEEKDAY-keyed
+    ({!Rite_of.Temporal_of.christmas_feria_slug}/the Advent ferial
+    branch), so looking one of THOSE slugs up would silently serve
+    whichever OTHER date happens to share that year's weekday alignment
+    -- the defect this function exists to close (bug found by review, not
+    by a test: colitur used to serve 17 December the unrelated
+    "advent-3-mon" reading in years where the two happened to align).
+
+    6 January is deliberately NOT one of the six Christmas-season dates,
+    despite lectio's own "christmas-jan-6" base existing:
+    {!Rite_of.Temporal_of}'s own [named] fixes Epiphany to 6 January
+    unconditionally (temporal_of.ml, "m = 1 && dd = 6"), so that date is
+    never reached as a ferial by {!readings}'s own step 3 in the first
+    place -- a real lectio entry with no reachable colitur day, the same
+    "structurally unreachable" shape as tools/bootstrap_lectionary_of
+    .ml's own "easter-6-thu"/"advent-4-sat" (see its own [excluded_bases]
+    comment).
+
+    [None] on a Sunday: Advent 4, Holy Family, the Second Sunday after
+    Christmas and the Baptism of the Lord can each land inside these two
+    windows, each is a NAMED office with its own distinct slug
+    ({!Rite_of.Temporal_of.sunday_slug}), and each must keep taking
+    priority over a ferial date -- checked here explicitly rather than
+    trusted to the call site, since this function's whole contract is
+    "this civil date, unconditionally, is this citation", which is false
+    on the Sunday the window shares its calendar date with. *)
+val date_keyed_slug : Date.t -> Slug.t option