perf: apply init-and-skip pattern to ciou and diou distance

[Smirkey]

Summary

Follow-up to #85, applying the same init-and-skip pattern that gcanat used on iou_distance_slice to ciou_distance_slice and diou_distance_slice.

The pattern: pre-fill result with utils::ONE once, then continue on the no-overlap branch — instead of initializing to zeros and explicitly writing result[idx] = ONE per non-overlapping pair. One fewer write + one fewer index calculation in the no-overlap hot path.

Also adds ciou_distance_benchmark and diou_distance_benchmark (neither existed in bench_iou.rs before) so CodSpeed can measure the delta on this PR and on any future change to these functions.

Other distance functions are not eligible for this pattern:

• giou_distance_slice always computes the enclosing-box term and writes every cell unconditionally.
• tiou_distance_slice has no no-overlap branch at all.

Test plan

• cd powerboxesrs && cargo test — all 91 + 18 doctests pass.
• cd powerboxesrs && cargo test --no-default-features — slice-only build, 52 + 2 doctests pass.
• cargo bench --bench bench_iou -- 'ciou distance|diou distance' --quick — new benches run (ciou ~50 µs, diou ~30 µs on 100×100).
• cargo fmt --check on both crates.
• CodSpeed: expect ciou distance benchmark and diou distance benchmark to appear in the report. If they show a regression or ~0% change, the LLVM optimizer was already eliding the redundant write — the refactor still stands as a readability cleanup.

[codspeed-hq]

Merging this PR will improve performance by 34.89%

⚡ 2 improved benchmarks
✅ 173 untouched benchmarks
🆕 2 new benchmarks
⏩ 1 skipped benchmark¹

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
⚡	test_diou_distance[float32]	212.6 µs	157.9 µs	+34.68%
⚡	test_diou_distance[float64]	224.3 µs	166.1 µs	+35.1%
🆕	ciou distance benchmark	N/A	1.5 ms	N/A
🆕	diou distance benchmark	N/A	257.4 µs	N/A

Tip

Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing perf/ciou-diou-init-skip (bb7d473) with main (6f1c650)}

Footnotes

1. 1 benchmark was skipped, so the baseline result was used instead. If it was deleted from the codebase, click here and archive it to remove it from the performance reports. ↩

[gcanat]

Guess you can skip that as well, boxes1 is already initialized to zero.

[gcanat]

same here