feat: add saturating quantize_round

[swgiacomelli]

Summary

• add const quantize_round:: with half-bin round-to-nearest, round-up ties, and a load-bearing full-scale clamp
• exhaustively verify correctness, excursion, range, full-scale saturation, 16-bit identity, tie direction, and consistency with quantize across every valid width and u16 input
• document the zero-parameter admission rule and keep parameterized dithering policy out of scope
• exercise the new API in the embedded link binary so resource and panic checks cover it

Verification

cargo xtask ci passed. All runnable checks passed; no checks skipped.

The final thumbv6m linked exercise sizes compared with pre-change main are:

• default: 2234 bytes, +84 bytes (limit 3072)
• f32: 5954 bytes, +92 bytes (limit 8192)
• oklab: 3550 bytes, +72 bytes (limit 5120)
• f32+oklab: 7270 bytes, +88 bytes (limit 9216)

Static RAM remains zero and no panic or wide-division symbols were linked.

Scope audit

No golden file, budget, existing quantize/expand behavior, type-state rule, dependency, version, tag, release, registry setting, or repository setting changed. ph-color-bake is untouched.

The brief did not conflict with AGENTS.md, docs/NUMERICS.md, CONTRIBUTING.md, or RELEASING.md. CONTRIBUTING.md additionally requires every new feature surface to be reachable from ph-color-exercise; that coupled edit is included so the resource evidence covers this API.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 15eed8ddbe

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".