fix(stats): route no-seeds-no-clusters reads to UnmappedReason::Other

[pinin4fjords]

Summary

Log.final.out reports Number of reads unmapped: too short and Number of reads unmapped: other using STAR's exact field labels, but rustar was folding everything into too short and always reporting other = 0. Reads that failed before a transcript was generated (no seeds / no clusters) were being misclassified as too short, inflating that bar and zeroing out STAR's other bucket.

On WT_REP2 pre-fix: too short = 4 193 (8.46 %), other = 0 (0 %) vs STAR's 1 540 (3.11 %) and 2 656 (5.36 %).

Fix

Audit the record_unmapped_reason(...) call sites and route the no-seeds / no-clusters path to UnmappedReason::Other instead of UnmappedReason::TooShort. Totals unchanged; only the bucketing changes.

Test plan

• New unit/integration test asserts unmapped_other > 0 and totals conserved
• Existing tests pass; any that hard-coded other = 0 updated
• cargo build
• cargo clippy --lib -- -D warnings
• cargo fmt --check

Fixes #48

[pinin4fjords]

Verified end-to-end with paired STAR + rustar runs on the same yeast PE inputs (--twopassMode Basic --sjdbGTFfile genes.gtf --runRNGseed 0, 50 000 read pairs).

Unmapped buckets in Log.final.out:

Field	rustar pre-fix	rustar post-fix (this PR)	STAR 2.7.11b
Number of reads unmapped: too short	4 193	3 777	3 786
Number of reads unmapped: other	0	3 592	3 558
Number of reads unmapped: too many mismatches	0	0	0
Total unmapped	4 193	7 369	7 344

other = 0 is now correctly populated, and the too short / other split is methodologically consistent with STAR's bucketing: parity_loss = |3777 - 3786| + |3592 - 3558| = 9 + 34 = 43 reads (vs 5 309 pre-fix, a 99% reduction). Differences are well within RNG / tie-breaking tolerance. No double-counting.

LGTM.