Re-shoot the README against the console that actually ships

[AndrewCTF]

Every still in the README was taken between 2026-07-10 and 2026-07-27, before
the console rebuild landed. They show the old shell: the flat icon rail, the
old palette, panels that no longer exist at those addresses. The first thing a
visitor sees was a product that is no longer there.

Twelve stills and the hero GIF are re-shot at 1440p against live feeds, and two
scripts make the set repeatable rather than something done by hand once.

What is here

• scripts/screenshot-readme.mjs drives each surface by a named recipe (app,
  camera, layers, tab, sub-tab, selection) and writes to the paths the README
  already embeds. Re-shooting the whole set is one command; --only NAME
  re-shoots one.
• scripts/record-hero-gif.mjs records the four beats the hero caption claims,
  in one take: the live world, the fly-in to Europe, an aircraft selected with
  its owned track and dossier, then the hour rewound on the replay dock. It
  encodes with a per-clip palette, which is what keeps a dark globe from
  banding into mud. 7.7 MB to 2.2 MB.
• The tests badge and the pytest comment said 1972. The suite is 2255.
• The airspace figure's alt text claimed a seaport dossier it never showed; it
  now describes the TFR polygons and hazard layers that are in the frame.

Traps the harness had to work around

Recorded in the scripts so the next person does not rediscover them:

• The low-end-graphics, degraded-feed and open-mode banners are notes about the
  capture box, not the product, and they ate two rows of every frame. All three
  are localStorage-dismissible; the harness sets those keys up front.
• Playwright's actionability wait never settles on this app, because the live
  feed mutates the DOM every second. A native el.click() still bubbles an
  event React's delegated handler catches.
• The ms-playwright cache on this box carries no chromium bundle, so both
  scripts pass an explicit executablePath.
• The time dock's Rewind sets the playback multiplier; it does not seek.
  Using it left the replay figure on live, the one state a replay figure must
  not be in. Jump to start is the seek.
• Info is four sections behind sub-tabs, and Workflows opens on an empty
  canvas, so both needed a click before there was anything to photograph.

Not re-shot, and why

ui-detach-toolbar.jpeg. The detach control lives in ConsoleShell, which
only the /2d route renders; the 3D console renders Console.tsx, which has
no detach affordance at all. README §11 ("tear the workspace apart") describes
something the shipping console cannot do. The same applies to the replay
figure's caption, which names a coverage chip ("recording since … · GB ·
fixes") belonging to Timeline, also 2D-only.

Both want a follow-up decision: restore those surfaces to the 3D console, or
rewrite the two paragraphs. Shooting them against /2d would drop a
different-looking console into the middle of a tour of this one, so the old
figure stays until that is decided.

Verification

bash scripts/verify.sh green: typecheck, lint, web unit, and 2255 passed +
2 skipped on the backend suite. Every figure was opened and read before being
committed, not just written to disk.