Adopt @statewalker/webrun-http-browser 0.5.0; mountEdge relies on its uncontrolled-page recovery

[mkotelnikov]

What

• Bump @statewalker/webrun-http-browser to ^0.5.0 (webrun-wire feat!: the membership token travels in x-httpeers-token, never Authorization #9, merge 485ad21) in httpeers-member (peer + dev), httpeers-browser-conformance, demos and llm-chat. A caret on 0.x pins the minor, so ^0.4.x would never have picked it up. The standalone lockfile changes only those edges.
• mountEdge relies on the library. 0.5's SwHttpAdapter.start() sends a CLAIM request that the stock sw-worker (every app's /sw.js) answers with clients.claim(), bounds every wait by timeout, and rejects with ServiceWorkerControlError. So the page-side pre-check and reload guard from fix(member): mountEdge recovers a page its ServiceWorker does not control (hard reload hang) #5 go; a hard-reloaded page is now taken over in place, with no navigation.
• edge-control.ts is now just edgeStartError: ServiceWorkerControlError (matched on name/reason, since every bundle carries its own copy of the class) → UncontrolledPageError ("close the tab and reopen it") for uncontrolled, the existing actionable Error for activation-timeout / unresponsive, anything else unchanged. The library error is kept as cause.
• reloadIfUncontrolled: true stays, as the last resort. It covers a worker that does not answer CLAIM: a pre-0.5 /sw.js still installed in a returning visitor's browser, or a custom serviceWorkerUrl. It costs nothing when the claim works, and the library guards it with sessionStorage so it can't loop. The catch is that such a worker keeps the page waiting for the whole controlTimeoutMs (30 s) before the reload.
• Public exports (mountEdge, UncontrolledPageError, DEFAULT_CONTROL_TIMEOUT_MS, controlTimeoutMs) are unchanged.
• No demos asset workaround exists on main (it lives only in unmerged Session origins: {name}.p.httpeers.net shell, demo, infra #3), so there is nothing to remove. The apps copy webrun-http-browser/sw-worker to sw.js at build time, so a rebuild ships the CLAIM handler.

Tests

• Standalone, as CI runs it: pnpm install --frozen-lockfile, lint:check, build, typecheck and test all pass (member 116, llm-chat 119, hub 125, access 185, …).
• In the llm-chat assembly (webrun-wire at main, every webrun-http-browser edge link:): the same suite passes.
• test:reload (Playwright, Chromium + Firefox), both linked and against the published 0.5.0 tarball:
  • first visit, normal reload, and a second tab (plus the first tab after it) all pass;
  • hard reload recovers in place: controlledAtLoad=false, one load, 1 navigation;
  • on a second origin whose worker ignores CLAIM: a hard reload does the one fallback reload (2 loads, 2 navigations, controlled). With the guard already spent it shows "close the tab and reopen it", with no reload. The normal reload after that recovers.
• test:browser: 11/11 (Chromium).

🤖 Generated with Claude Code