A browser that dies or freezes without closing its websocket used to hang the run forever: the node-side rpc has no timeout, and every existing liveness check (testTimeout, iframe ack) runs inside the browser itself, so none of them can fire once the browser process is gone.

• the server now pings every browser connection and terminates it when pings go unanswered, so the pending createTesters call rejects and the run fails with an actionable error. Browsers answer these pings from the network process, not from page JavaScript, so a sync-busy test or a page paused in the debugger does not false-positive. VITEST_BROWSER_HEARTBEAT_INTERVAL overrides the interval (0 disables it); an invalid value prints a warning and falls back to the default instead of silently disabling the heartbeat.
• the browser pool no longer waits forever for provider.close(): an unresponsive browser previously blocked close() and the process could never exit. The browser process is killed by the provider when the process exits.
• the playwright provider subscribes to page.on('crash') again (the handler was removed in perf(browser): improve browser parallelisation #7665) and fails the run immediately with an attributed error. Without it, a crash is only visible as a websocket disconnect, if the browser closes the socket at all.

Closes #10791
Closes #10151