colitur 1.2.1 -- ten wrong reading references, corrected

A correctness release for the Ordinary Form lectionary. Nothing about the
engine changed and no output format moved; ten scripture references did.

Nine ended at verse 22 where the OLM ends them at 20 -- Mark 1:14, Mark 4:1,
Mark 5:1, Matthew 7:15, Matthew 17:14, Matthew 18:15, 1 Samuel 1:9,
1 Samuel 3:1-10.19, James 5:13. The tenth read "Matthew 17:9a,10-135", a verse
Matthew 17 does not have.

Every one sat in a Year II entry, or in of-advent-2-saturday, which has no
year cycle. Not one Year I entry was affected, and that distribution is what
makes this a single transcription fault rather than ten independent slips.

Eight corrections are confirmed verbatim in the Latin OLM 1981 -- the primary
source, not a witness. The two its OCR does not yield are impossible on their
face: James ends at 5:20, and Matthew 17 has twenty-seven verses.

The defect was upstream, in lectio, and copied faithfully here. It is fixed
there and this calendar regenerated from the corrected source, per the data
file's own instruction never to hand-edit a generated artifact; the diff is
exactly the source SHA-256 pin plus the ten references.

WHY IT SURVIVED THREE RELEASES. The OF's validation compared season, grade and
identity, and stopped. Every reading it served had been checked against
nothing but the lineage it arrived on. This release adds the first readings
validation, and it found all ten on its first run.

Two detections, and neither alone is sufficient. An internal invariant needing
no external data: in Ordinary Time the OF weekday Gospel does not depend on
the year of the cycle -- OLM n. 69 point 4 alternates only the First reading --
so a -i/-ii pair whose Gospel differs is a defect by construction. That catches
six. A comparison against litcal catches three of those, plus the four the
invariant structurally cannot see: three First readings and an Advent entry,
which no such symmetry constrains.

The invariant is scoped to Ordinary Time deliberately. Unscoped it reports
seventeen further pairs across Advent 4 and the Christmas octave, every one of
them correct, because those weekdays are keyed to the calendar date rather than
to the cycle. An unscoped version would have been switched off by whoever read
it next, and rightly.

Measured: the 20-vs-22 divergence class against litcal goes 26 to 0 over
twelve witnessed years.

Also in this release: --help is 42 lines rather than 374, and points at
colitur(1) for everything it no longer repeats. Each dropped section was
checked to have a man-page counterpart. One test moved rather than died --
cli.t pinned rubrics' documented field list against what rubrics actually
prints, because that count drifted once and stayed wrong for a whole review
round; it now reads the worked example out of the man page.

917 tests, every year 1583-9999.