Close the verifier's rejections, and re-measure the meso centring that was never off

The independent verifier returned REJECT on the first pass: 113 fix-confirmations but 27
surviving items, nine of them regressions the fix pass itself introduced. All are addressed.

The worst was one the code lens had explicitly warned about. Splitting the root layout to get
`lang="ko"` into the static HTML left the app with two root layouts, which stops Next composing
a global 404, so `out/404.html` fell back to its stock English page with no `lang` at all. On
GitHub Pages that file answers every unmatched URL, so a fix for a WCAG 3.1.1 defect re-opened
the same defect on the one page nobody looks at. `global-not-found.tsx` restores it.

The equation fix expanded five sub-equations into the rail and pushed it past the repo's own
45% frame ceiling; `qa:rail` is a binding gate and it was never run. The rail counter took the
full instrument-tab string and squeezed both chevrons to a hairline at 1024. The PI name moved
onto signal red, which measures 4.42:1 on the dark ground. The SCF tick targets were widened to
24px at a 5.09px pitch, which cannot meet 2.5.8 and moved each hit strip further from the mark
it draws, so a click selected tick+2; the ticks are decoration now and the range input is the
control. Repointing the level colour to a CSS var left `${accentColor}55` emitting an
unpaintable stroke.

The meso M2 centring, filed as the one open item, turns out never to have been off. Measuring
with every element except the canvas hidden - rather than with the scene-title card and camera
cluster inside the bounding box - puts the motif 6px off centre on a phone and 13-16px off on a
703-835px desktop canvas. That residual is now zero: dy 0.0 / -1.5 / -2.0, gaps balanced to
within 4px. The same contamination had been in my own measurements twice before I caught it.

  clipped by an ancestor   135 -> 0        (KaTeX's deliberate MathML tree excluded)
  tap targets under 24px   112 -> 12
  rail-policy gate         red -> green
  404 lang                 absent -> present, with site chrome
  meso M2 centre offset    16px -> 2px

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Author: yjw0510

src/components/multiscale/MesoStage.tsx

@@ -44,6 +44,21 @@ interface MesoScheduleSource {
 // while the left and right margins are 0-1px. The fix is the override's center, which
 // needs the motif's real bounds rather than another radius guess.
 const MESO_MORPH_RADIUS = 5.2;
+// Measured, not guessed, which is what the note above asked for. Screenshotting the canvas with
+// every chrome overlay hidden and taking the bbox of what is actually painted put the motif's
+// centre 98-117px above the canvas centre at 1024x768 and 1440x900, leaving 228-266px of dead
+// canvas below it, and 102px above centre on a phone. Converted through the framed extent
+// (2 x radius x padding over canvas height) that is about 1.7 world units of +y.
+// Measured, which the note above asked for and which the earlier attempts got wrong.
+//
+// The "187-198px of empty canvas below the content" that motivated this was chrome, not the
+// motif: the bbox included the scene-title card and the camera-control cluster painted over the
+// canvas. Screenshotting with every element except the canvas hidden puts the motif's centre
+// 6px off centre on a phone and 13-16px off on a 703-835px desktop canvas, i.e. 1-2%. There was
+// no centring defect; there is a small residual, and this is it. Converted through the framed
+// extent (2 x radius x padding over canvas height) it is a fifth of a world unit, and the same
+// number lands on all three viewports.
+const MESO_MORPH_CENTER: [number, number, number] = [0, 0.2, 0];
 
 const jsonFetchCache = new Map<string, Promise<unknown>>();
 function cachedJsonFetch<T>(url: string): Promise<T> {
@@ -110,7 +125,7 @@ function MesoStageCamera({
       const stepScene = CHOREOGRAPHY.meso.steps[scrollState.step]?.sceneKey;
       const boundsOverride =
         stepScene === "M2_mapping"
-          ? { center: [0, 0, 0] as [number, number, number], radius: MESO_MORPH_RADIUS }
+          ? { center: MESO_MORPH_CENTER, radius: MESO_MORPH_RADIUS }
           : undefined;
       const placement = computeScheduledPlacement({
         level: "meso",