lody: mount the real mobile branch, scrubbed to the v1 scope — seam patch 16

[pythonlearner1025]

The diagnosis: the test, not the poller

packages/webapp/test/shell-gateway-resilience.test.tsx >
the platform snapshot poller >
never has two reads in flight against a box that does not answer
timed out at 5 s on clean main (3881574).

The poller is correct. useLodyPlatformSnapshot in
packages/webapp/src/lody/platform.tsx has not changed since #155 landed it.
git log names one commit for that file: 6a6cb18d (#155). None of #156-#164,
#172 or the cgroup changes touched it. The single-flight invariant holds. I
measured it: with a fetch that never answers, the hook issues exactly one
read, and 30 000 ms of fake time add none.

The test's timed region was wrong. The test imported
../src/lody/platform.js inside its own it() body. That module statically
pulls convex/react, @lody/platform and two @lody/components providers, so
the import is seconds of transform. Instrumented on this box with a real clock:

start:                    +0ms
A fake timers on:         +9ms
B imported platform.js:   +7720ms      <- the whole budget, spent here
C rendered reads=1:       +7765ms
D advanced reads=1:       +7766ms      <- the property itself: 1ms

The 5 s test timeout fired on module weight. The property was never reached.
The other ten tests in the file pass because their heavy import
(SessionSurface) is lazy() and mocked away.

The fix

The import moves to the file's static import block. The transform then lands in
the file's import phase, which no test timeout governs.

The assertion is unchanged. expect(reads).toBe(1) after
advanceTimersByTimeAsync(30_000) against a hanging fetch still states the
single-flight invariant, word for word. Nothing is relaxed and no test is
removed. The test now takes 27 ms instead of timing out at 5 018 ms.

Gates

Gate	Result
npm run typecheck	pass
npm run lint:gate	pass — 74 anti-slop findings, all at baseline; 0 blitz-house
BLITZDEV_MANAGED=1 npm test — webapp	97 files, 857 passed, 6 skipped, 0 failed
BLITZDEV_MANAGED=1 npm test — box guest-tests	17 files, 106 passed, 0 failed
control-plane vitest	66 files, 723 passed, 0 failed
control-plane test:python / test:bash / test:host / test:scripts	1 / 4 / 13 / 59 pass, 0 fail

The control-plane stages were rerun alone: the box was running another agent's
full webapp suite at the same time, and the first combined run failed under that
contention. Every stage passes on its own with zero assertion failures.

The target suite was also run five times in a row after the fix: 11 passed,
five times out of five.