fix(sam): paired-end nM is the pair's mismatch total, as in STAR

[BenjaminDEMAILLE]

Follow-up to #293, where this divergence came up.

Problem

For a pair, STAR writes nM per pair, as it does for AS: both mates carry the sum of the two mates' mismatches. rustar-aligner wrote each mate's own count:

rustar  99  chr21 23390830 1S73M1S ... AS:i:128 nM:i:5
rustar  163 chr21 23390779 75M     ... AS:i:128 nM:i:4
STAR    99  chr21 23390830 1S73M1S ... AS:i:128 nM:i:9
STAR    163 chr21 23390779 75M     ... AS:i:128 nM:i:9

Change

build_paired_mate_record (src/io/sam.rs) writes transcript.n_mismatch + mate_transcript.n_mismatch for nM. Nothing else changes:

• NM is still per mate. It is an edit distance computed from the mate's own CIGAR, and STAR computes it per mate as well.
• Half-mapped mates (built in another function) keep their own count.
• The per-mate n_mismatch values on the transcripts are untouched, so the mismatch filters and fix(quant): filter TranscriptomeSAM pairs as a unit, like STAR (#292) #293's pair budget still see per-mate counts.

Results

Same data as #293: the first 2M pairs of ERR188021 against chr21 + GENCODE v44 chr21, with a STAR 2.7.11b index. Each paired record is matched to STAR's on (read, mate, position, CIGAR, mate position), secondaries included:

	nM equal to STAR	nM differs
main	633,910	247,069
this PR	880,623	356

I checked a sample of the 356 that still differ. In each case the other mate aligns differently (for example, it is spliced in STAR and not in rustar), so the pair's mismatch count legitimately differs.

Tests

• test_build_paired_mate_record_per_mate_tags and test_build_paired_mate_record_mate_fields asserted the old per-mate nM; they now assert the pair total. Mate1 in the first test now has 1 mismatch so that the sum differs from either mate's own count.
• cargo test --release and cargo clippy --all-targets pass.

🤖 Generated with Claude Code

[Psy-Fer]

PE nM = pair total landed in #307