[handrews]

data/ranks.yaml is now the rank vocabulary: the formal ranks in order,
as the data writes them, and the informal words flagged as such; it also
holds the suffix lint's rules (the ending a rank expects, the ranks an
ending expects, the year from which a name is checked) and the names the
lint lets pass, each with a note. phylohist/ranks.py reads it at import
for the constants both sides use, a load reads its own copy into a
RankTable the corpus carries, the schema's rank enum is held equal to
it by a test, and the loader errors on an exception that names no record
or a record at another rank. The old _check_rank, with its two loops,
46 inline exception names and a skip of every rank variant by its key,
goes: the lint checks records above the genus group from 1900 on, in
both directions, passes over another spelling, a gender form and a
vernacular form (a new Kind.suffix_checked flag), and checks a Latin
spelling and a rank variant at their rank. A superfamily accepts -acea
before 1999 and -oidea always; -oidea itself implies no rank, since
echinoderm classes end in it. Seven exceptions survive the cut in
data/, one more in the frozen test corpus, which has its own copy.

A record's rank: null means the source prints no rank word and no
cladistic term, and the code now reads it as none: the loader no longer
substitutes Unranked, which leaves the vocabulary (a modern unranked
clade is Clade), and the three unnamed Müller 1854 groups written with
it become null. The claims and the names index carry rank: null for
those records and their spellings (24 claim lines, 10 rows), headings
and labels print the name without a rank word, a placeholder still reads
"Unnamed unranked", and resolve_name(rank='unranked') finds them. The
open-nomenclature checks treat a record with no rank as unknown, and a
mineral, which has none, as no species-level name.