Charts: honour prefers-reduced-motion across every Recharts series

[aurph]

Two bugs, one cause

Recharts animates every series by default and never consults the user's motion preference. All 27 series on the site grew from zero regardless of what the visitor asked their OS for. That is the accessibility bug.

The second effect is subtler and is why the compute-page charts could never be verified. Recharts builds series geometry inside react-smooth, which only emits the actual <path> on its first requestAnimationFrame tick. A chart that never receives a frame renders structurally complete but visually empty — the <g class="recharts-bar-rectangle"> groups are all present with nothing inside:

<g class="recharts-layer recharts-bar-rectangles">
  <g class="recharts-layer recharts-bar-rectangle"></g>   <!-- no path -->
  <g class="recharts-layer recharts-bar-rectangle"></g>   <!-- no path -->

Nothing errors. Nothing logs. The chart just looks blank. Any visitor whose rAF is throttled hits the same thing.

Change

• chart-theme.ts exports prefersReducedMotion and seriesAnimation, following the existing axisProps/gridProps convention of a plain spreadable const.
• Every Recharts <Bar>, <Line>, <Area>, <Pie>, <Radar> spreads it — 27 series across 9 files.
• @visx deliberately excluded. PriceHistoryChart's <Area> comes from @visx/shape, a pure path generator with no animation and no such prop. Passing it would be meaningless, so the guard test skips files that don't import from recharts.

The guard test

The failure mode is a silently blank chart, not an error, so a new chart added without the prop would regress this invisibly. chart-motion.test.ts walks client/src and fails listing any offender. It found 9 series I missed on the first pass (PortfolioOverlay, TiltOverview) plus correctly flagged the visx false positive, which is what scoped the rule.

Verification

Run with chrome --force-prefers-reduced-motion, the Compute Frontier operator bars render real geometry, and the widths match the API to within rounding:

operator	bar px	API MW	bar ratio	data ratio
Meta	132	16,245	1.00	1.00
Amazon (AWS)	88	11,043	0.67	0.68
Google	83	10,460	0.63	0.64
OpenAI / Oracle	71	8,902	0.54	0.55

This closes the "charts unverifiable in headless" gap I flagged on #12/#13 — they were correct all along, just unprovable.

• npm test 303 pass, 0 fail
• npx tsc --noEmit clean
• npm run build passes

Independent of #12 and #13; touches no files either of them touch.

[aurph]

Verified against current main (e887487) in a clean worktree:

• tsc clean
• 290 tests pass (287 baseline + 3 new)
• build passes
• Merges clean onto main, and clean alongside Social: rebuild the daily card with real type, real map, and provenance #12 and Home: put a refresh date on every front-page figure #13

Confirmed the second effect you describe: on the merged branch the compute-frontier bar and timeline charts render with visible geometry in a headless Chrome screenshot. The source-walking test is the valuable part here, since it fails on any future series added without the prop rather than letting the chart go silently blank.