Test that browser nonzero exits do not alter guest stderr

[brandonpayton]

What #907 already landed

#907 landed the browser runtime behavior that keeps host diagnostics out of guest stderr when a process exits nonzero.

What remains here

This PR keeps one Chromium regression. It resolves the maintained nonzero-exit fixture through resolveBinary, runs it in the browser, and checks that the real guest stderr is unchanged.

What I removed

I removed the duplicate runtime implementation and all #889 ancestry. The rewritten branch is one test-only commit directly on current main.

ABI and package artifacts

There is no ABI change. No package recipe, package revision, VFS image, or published artifact changes.

Validation

Run through scripts/dev-shell.sh:

• host TypeScript typecheck: passed
• browser asset check: passed (92 imports resolved)
• focused Chromium regression: passed
• Chromium wait-lifecycle controls for wasm32 and wasm64: 2 passed
• full non-slow Chromium suite after preparing browser assets: 34 passed, 4 skipped
• ./run.sh browser: built the browser images and started Vite successfully; /pages/kandelo/ loaded in Chromium and the running shell demo was visually checked
• full host Vitest attempt: 1,322 passed, 42 skipped, 2 expected failures; three package-contract files failed because local Cargo archiving returned status 255

Not run locally: the Firefox/WebKit cross-browser smoke. The PR CI run is the authoritative full-suite check.

[brandonpayton]

CI note: the failing test-gate-prepare did not reach this PRs Playwright regression. The separate Browser demo smoke workflow passed. Fetch-only materialization rejected four ABI-16 composite packages because the durable release index.toml currently points at older keys (lamp 9f777974, node-vfs 5ca2294d, shell 2f4b693f, wordpress 6c52e8a5) while current main computes the newer #863-promoted keys (64104909, 7618c9c6, f7dccdb9, 76133d30). Both generations of archives are present in the release; the index was updated later than the newer assets. I am leaving this host-only patch unchanged and will rerun the gate after #867 publishes the ABI-18 baseline.

[brandonpayton]

Rewrite record: before removing the stale #889 ancestry, this PR pointed at ed3f2afae64d260d906f7d712b21d8eb558998a8. I am reconstructing its unique Chromium regression directly on current main; the runtime change is already on main through #907.

[brandonpayton]

Closing this focused source PR because #934 now carries the unique Chromium nonzero-exit stderr regression at exact green head 260d75d5adc31b327f5591229dbfbb287d395fd6. The runtime change already on main through #907 was deliberately excluded. The full staging and browser gates passed. This is not a claim that this branch was merged verbatim.