Strict colon loci: colon keys, peaks converted at import

[YPARK]

Loci are recognized only in colon form, chr:start-end, through legume-genomic-types 0.5.0. Outside spellings are converted once where names enter, and everything inside works on colon keys.

Keys

• Canonical key {chr}:{start}-{end}: chr prefix dropped in any case, the rest's case kept (chrX:0-100, CHRX:0-100 and X:0-100 all give X:0-100).
• feature_names::locus_key (re-exported from genomic-types) is the one answer to "is this a locus, and under which key". The overlap map and every per-name path now key a row the same way.

Matching

• The gene rule never cuts a locus (tagged with a feature-type suffix or not) and keeps names whose symbol would be all digits (chr1_100_200), so such rows cannot collapse onto one key.
• auto_detect warns when an axis reads as intervals only in a non-colon spelling.
• GeneIndex resolves a locus query only against locus rows by key, case kept, including rows whose /-core is a locus. Gene queries never land on locus rows.

Import boundary

• colon_peak_names rewrites peak rows to chr:start-end (from chr1-100-200, chr1_100_200, chr1:100_200). Rows count as peaks by feature type (peak / ATAC) for 10x h5/zarr, features.tsv and h5ad, or for an untyped list when every row reads as an interval.

Breaking (0.7.0)

• Keys change from X_0_100 to X:0-100. Names like chr1_100_200 stored in existing backends are no longer loci.
• parse_locus accepts colon form only.

Version 0.7.0.