test(ef): compare the LMS Ordo's formulary override STRING, not just presence

test_lms_ordo.ml's formulary-override check only asked WHETHER a
Preceding_sunday day showed a "Mass of ..." override line, never WHICH
Sunday it named. colitur could have resumed the wrong Sunday on all
66 such days and this layer would have stayed green -- resuming the
right Sunday is precisely what step 3 of the reading chain exists to
do.

test_formulary_override_matches now derives the Ordo's own expected
prose from Mass_formulary.t.said's own resolved slug (parsed directly,
no Str/regex) and compares it against the Ordo's real text. Two
narrow, evidenced exceptions, both counted rather than silently
absorbed:

  - Christ the King's own Sunday reaches this population under a
    NAMED slug (ef-christ-the-king) that carries no embedded ordinal;
    its position among the Sundays after Pentecost varies by year, so
    Temporal_ef.week is read for that one slug specifically, not
    called generically (an earlier version of this fix read [week]
    for every day and was wrong -- caught by its own assertion on the
    very first run, because Time_after_pentecost's own last-Sunday and
    surplus-Sunday renumbering diverge from the raw week count. Reading
    the number back out of the already-resolved slug sidesteps both
    special cases and is not tautological with the wrong-Sunday
    failure mode this fix exists to catch).

  - Three real days between Ascension and the following Sunday
    (2025-05-30, 2025-06-02, 2025-06-03) are a genuine, newly-found
    defect: colitur resumes the nearest calendar Sunday, but the
    Missal, and the Ordo, keep saying Ascension's own Mass through
    this stretch instead. Structural (Ascension is a fixed Easter+39),
    not a fixture artefact, and not fixed here -- it needs a
    lectionary-data change out of this fix round's own scope.
    Documented in expected-divergences-lms.sexp as a prose-only note
    (no active sexp record: this file's id-lookup machinery is scoped
    to the Creed axis, and a second axis reusing it would let one
    test's "declared but never fired" check spuriously trip on the
    other's entry) and enforced directly by this test's own count
    assertion instead.

Mutation-proved: offsetting step 3's Sunday lookup by an extra 7 days
reddens this test (23 failures total across the suite: layers 3-6,
golden pins, and lectionary_ef's own unit tests), confirmed via the
compiled test binary directly; reverted clean, full suite green again.

0 files changed, 0 insertions, 0 deletions