perf(align): memoize the junction-position scan inside stitchWindowAligns

[BenjaminDEMAILLE]

What

find_best_junction_position is the single hottest function in the aligner. On a 50k-pair yeast PE run (sample-based profile, single thread) it is ~60% of alignment time, inside stitch_seeds_core at ~76%.

The scan itself is not wasteful, but it is repeated. stitchWindowAligns' include/exclude recursion reaches the same (exon A end, seed B) pair through many different branch paths, and every path re-runs the identical scan.

The scan is a pure function of its arguments. Within one window read_seq, the genome, is_reverse and n_genome are fixed, so six coordinates identify a scan completely: exon A read end, exon A genome end, read gap, genome gap, previous exon length, next seed length.

This PR adds JunctionScanCache (a per-window FxHashMap on that key) and a find_best_junction_position_cached wrapper. The uncached function is untouched, so a cache hit returns exactly what a fresh scan would have returned. The cache is created per window in stitch_seeds_core and threaded down the recursion, which is what keeps the key complete: it never outlives the read, genome and strand it was filled for. An empty HashMap does not allocate, so windows that stitch nothing pay nothing.

Effect on the profile

find_best_junction_position inclusive samples drop from 6,855 to 2,941, a 57% cut in the function's total cost.

Benchmark

50k yeast read pairs, Apple M4 Max, quiet machine, best-of-6 interleaved rounds, --outSAMtype None:

threads	before	after	change
1	19.51s	16.76s	-14.1%
8	2.55s	2.18s	-14.5%

Correctness

Output is byte-identical. Aligned.out.sam (records and header, modulo the @PG CL line naming the binary) and SJ.out.tab compare equal against the pre-change binary on the same input.

• 592 tests pass
• 0 clippy warnings (--all-targets --release)
• cargo fmt --check clean

🤖 Generated with Claude Code