Better handling of small rays

[facelessuser]

Do not discount an axis just because distance is very small, but instead reject an entire ray if all axes have very small distances.

Fixes issue, most often with very small rays near white and black, that exited improperly causing the colors to not be correct.

A note was added that an offset of 1e-5 for anchor adjustment may be preferable as real world tests have shone that math functions on various systems may be more coarse, and produce more noise than 1e-6 can handle. In tests, it seems this adjustment is only needed for the offset related to moving the anchor closer to the gamut surface. The cutoff for ray distance in raytrace_box is currently still recommended to be 1e-6.

Issues found by @lloydk

[netlify]

✅ Deploy Preview for colorjs ready!

Name	Link
🔨 Latest commit	1dafc62
🔍 Latest deploy log	https://app.netlify.com/projects/colorjs/deploys/6ad2687e7b4cafb02a1cf767
😎 Deploy Preview	https://deploy-preview-773--colorjs.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.
🤖 Make changes	Run an agent on this branch

---

To edit notification comments on pull requests, go to your Netlify project configuration.

[facelessuser]

For more context, see the discussion between @lloydk and me over at this ColorAide issue: facelessuser/coloraide#529.

[lloydk]

I had Claude review the PR:

I checked this against near-black, near-white and just-outside-gamut sweeps in sRGB, P3 and Rec.2020, with the cast instrumented to flag any ray that lands outside the RGB cube. On main, rays land outside the cube for 351–406 near-black inputs per gamut, and almost all of them come out visibly wrong: saturated red or magenta instead of black. Near white, main also casts some rays outside the cube, but none of those results were visibly wrong. With this PR, no ray lands outside the cube in any sweep. Over 1,000,000 random colors plus the near-white sweep, the PR changes 225–267 results per gamut, and none gets worse by more than 1e-15.

Only one of the new tests fails without the fix: oklch(0.0001220703125 0.4 19.75) in sRGB. The other five pass on main too. These near-black cases fail on main and pass with this PR:

// sRGB
{
        args: ["oklch(0.0000951210968196392 0.41658416169229895 328.082845620811)"],
        expect: "rgb(0% 0% 0%)",
},
// P3
{
        args: ["oklch(0.0001220703125 0.4 19.5)"],
        expect: "color(display-p3 0 0 0)",
},
// Rec. 2020
{
        args: ["oklch(0.0001220703125 0.4 17)"],
        expect: "color(rec2020 0 0 0)",
},

Two small things:

• The // 32-bit should use 1e-6 line is indented with spaces; the rest of the file uses tabs.
• "impelmentation" in the new anchor comment should be "implementation".

[facelessuser]

I've updated whitespace, spelling, and added the additional tests that target the Color.js implementation more specifically. As there are a lot of things that influence the noise between different implementations, it is good to have tests that specifically target this implementation.

[lloydk]

LGTM