Locus rows keep chromosome case; one grammar from legume-genomic-types

[YPARK]

Problem

parse_locus lowercased the whole row name, so auto-detected Locus { merge_overlapping: true } and Mixed alignment (e.g. read_data_on_shared_rows) renamed chrX:0-100 to x_0_100. Names outside the overlap map fell back to canon_locus, which keeps case (X_0_100), so one chromosome could come out two ways, and case-sensitive chromosome matching downstream silently missed X, Y and M.

The parser also split from the left, so contig names carrying _ or - stopped being loci. On a Mixed axis those then went through the gene rule, and distinct contig peaks could collapse into one row.

Change

The locus grammar and its canonical key now come from legume-genomic-types 0.4.6 (coordinates::parse_interval, PeakCoord::locus_key), the single owner of the grammar.

• Keys are {chr}_{start}_{end} with chr dropped and case kept, on every path: the overlap map, its fallback, Locus without merging, and Mixed.
• Contig names keep their _ and - and round-trip through the key.
• Mixed per-name dispatch lives in one helper shared by canonicalize and build_mixed_kind_canonicalizer.
• The fragments builder's chromosome key reuses chr_stripped instead of a local copy.
• parse_locus keeps its signature.

Behaviour changes (patch release 0.6.24)

• Letter chromosomes keep their case in canonical row names (x_… becomes X_…). Outputs or models stored with the old lowercase names will not match these rows by case-sensitive lookup.
• An empty interval (start == end) is no longer treated as a locus.
• A non-locus name on a Locus axis now passes through unchanged instead of being separator-folded.

Tests

• New: shared-rows load of two files keeps X_0_100 (fails before this change).
• New: case kept on every canonicalizer path; contig names with _ parse and round-trip; contig peaks stay distinct loci on a Mixed axis.