test: vitest suite — scales/palette units + a mount regression for #2

[DloomPlz]

What

A zero-config test harness (vitest + happy-dom + @testing-library/react, all devDependencies — no changes to registry sources) with three suites:

• tests/scales.test.ts — the pure geometry: computeBands across default/stacked/percent (including dirty input coercion and the empty-data guard), buildYScale, the x scales, and pixel hit-testing (nearestIndex, indexAtBand).
• tests/palette.test.ts — seed shape invariants, rgb() brightness/alpha math, isDitherColor guards.
• tests/chart-mount.test.tsx — a regression for Infinite "Maximum update depth exceeded" loop in all charts without React Compiler #2: mount AreaChart/PieChart/Sparkline under a compiler-less React with a measured size (the loop only reproduces once the chart is ready) and assert no update-depth loop.

The mount tests are it.skip — deliberately

On current main the #2 loop doesn't fail in a test harness — it hangs the worker (the effect loop keeps the scheduler busy forever), which would hang CI rather than redden it. So they're skipped with instructions in the file: unskip them in the PR that merges #3/#4, at which point they pass and become the permanent regression guard.

Verified both directions locally: suite is green on main as-is (16 passed / 3 skipped, <1s), and with the #3 + #4 patches applied and the skips removed, all 19 pass.

npm test is wired up; happy to add a GitHub Actions workflow on top if you want CI in the same PR.

[tripwire-dev]

blocked by tripwire — the details are in the tripwire comment on this PR.

[tripwire-dev]

passed — nothing tripped. good to merge.

[DloomPlz]

Slimmed this down: dropped package-lock.json from the PR (and gitignored it) — it was 1,995 of the 2,275 added lines for zero consumer value in a registry repo. The hand-written change is now ~280 lines; suite still green locally (16 passed / 3 skipped, <1s).

Two questions for the tripwire run:

1. Which rule tripped? The run links in the bot comment point at http://localhost:3000/runs/…, which isn't reachable for contributors — looks like the instance's public base URL might be misconfigured. If the tripped rule was the skipped tests (it.skip on the three mount regressions), that's deliberate and documented in the file: on current main the Infinite "Maximum update depth exceeded" loop in all charts without React Compiler #2 bug hangs the test worker rather than failing, so running them would hang CI — they're meant to be unskipped in the PR that merges fix: stabilize register/unregister callbacks without React Compiler #3/fix: memoize Sparkline's derived rows/config (companion to #3, closes the rest of #2) #4. Happy to restructure (e.g. behind an env flag) if the rule can't be waived.
2. If it was diff size / lockfile, the push above should clear it on re-run.