bench: add a divan harness and the first hot-path benchmarks

[BenjaminDEMAILLE]

Closes #204.

Several open dependency questions (#162, #202, #205, #208) all start with "measure first", and each was about to invent its own measurement. This is the shared one.

Divan rather than criterion

Both were on the table in the issue. Divan because the questions that prompted a harness compare peak memory as much as speed (sufr/libsais against caps-sa is explicitly about genomeGenerate wall time or peak RSS), and divan reports allocation counts next to wall time with no extra setup. A full run also takes seconds, which keeps it usable inside a pull request rather than only in a nightly job.

Criterion is the better choice if the CI integration in #245 wants its report format. Swapping is a benches/ change, not an API one, so this does not foreclose it.

What is measured

Three groups, picked as the functions the open questions would actually replace:

• seed_scan_full_match / seed_scan_stop_at_half — simd_scan::find_stop, the innermost loop of seed extension and the function a portable-SIMD crate (Dependency discussion: portable SIMD crate (wide/pulp) vs the hand-rolled intrinsics in simd_scan.rs #205) would swap. Two shapes, because a full match and an early stop have different costs.
• gene_overlap_query — the segment-tree overlap query, per read and twice for GeneFull, and what an interval crate (Dependency discussion: interval-overlap crates (superintervals / coitrees / rust-lapper) #208) would swap. Sized at 100 / 2 000 / 20 000 genes.
• gene_annotation_build — the allocation-heavy half of the same question.

All are pure functions with no genome index to build, so a run takes seconds and a regression is attributable to one function rather than to a pipeline.

Sample output (--sample-count 3, an M-series laptop):

seed_scan_full_match     50    3.0 ns
                        250    8.1 ns
gene_overlap_query      100   101.5 ns
                      20000   189.1 ns

The bench feature

find_stop and the GTF record type are private to the crate. Rather than widen the published API for a benchmark, both are exposed under a bench feature that only benches/ enables, marked #[doc(hidden)]. Nothing changes for a normal build or for crates.io consumers.

The DEPENDENCIES.md row for divan belongs with #246, which introduces that file; the rationale above will move there rather than being duplicated.

cargo test, cargo clippy --all-targets -- -D warnings on a cold cache, and cargo fmt --check are green.

🤖 Generated with Claude Code