Carry the measured gene, the element's location, and the channel

[emattei]

Three gaps, each found by reading the assembled MuData and asking where
something had gone.

The measured gene, in the element mapping

guide_element_mapping.tsv named no gene, so the mapping could not stand on
its own. It now takes the same --genes pairs table igvf-export does,
through the same _measured_genes lookup, so the two tables cannot disagree
about which gene an element was measured against.

promoter        3,380 / 3,890
e2g_validated     180 / 379
peak                0 / 44,501

element_location

genomic_element cannot say whether an element sits in a gene: promoter and
enhancer win on precedence before the gene body is ever tested. The cross-tab
is the argument for a separate column.

                 genic   intergenic
distal element       0       14,900
exon             2,995            0
intron          14,631            0
promoter         8,348          499     <- split
enhancer           230          110     <- split

Computed from the same in_gene overlap igvf_export already had, so no new
input and no second source of truth.

What the channel is

The object said nothing about which channel it held. --channel NAME=VALUE,
repeatable, joins every cell's obs and is kept in uns['channel'], so
replicate, day, chip and batch travel with the counts.

Stated rather than parsed. The names live in FASTQ paths, and reading them
there would tie the tool to one lab's naming scheme.

730 tests pass.

🤖 Generated with Claude Code