fix: wait for the browser to actually exit when stopping it

[nathanfallet]

Problem

stop() asked the browser to terminate and moved on:

withTimeoutOrNull(5.seconds) { connection?.browser?.close() }
process?.destroy()
process = null

destroy() only requests termination — it returns immediately — and it does not reach the browser's child processes. On Windows the renderer and GPU children outlive it and keep open handles on the profile directory.

So a browser started on the same --user-data-dir shortly afterwards cannot take ownership of the profile: it hangs before opening its debug port, and the start times out after the full connect window with no visible cause.

We hit this in production as a restart loop: close the browser, reopen it one second later, fail for 30s, repeat — recoverable only by rebooting the machine, which is simply what finally kills the leftover children.

15:44:05  restart requested → stop() → destroy() (async)
15:44:06  new browser started on the same profile   ← one second later
15:44:53  failed: debug port never opened           ← 31s timeout

Fix

stop() now waits:

process?.let { if (!it.destroyAndAwaitExit()) logger.warn("Browser process ${it.pid()} still alive after kill") }

destroyAndAwaitExit() follows the same shape as zendriver's Browser.stop (terminate → poll up to 3s → kill → wait):

1. destroy(), then poll isAlive() for a grace period (3s by default);
2. still alive → killTree();
3. poll again (5s by default);
4. return whether it actually exited, so the caller can log if it did not.

The waiting loop is common code — isAlive() exists on every target — so only the forceful kill needed a per-platform actual:

• JVM: killTree() also takes down descendants. They are snapshot before killing the parent, since killing it detaches them and they can no longer be enumerated.
• POSIX (linux/macos): SIGKILL to the process itself, which is enough there — unlike Windows, children do not outlive their parent.

Notes

No API change beyond the new Process.killTree() expect/actual and the destroyAndAwaitExit() helper. Timeouts are parameters with sane defaults.

:core:jvmTest passes. :core:build fails on kotlinStoreYarnLock, as on main — unrelated.

[codecov]

Codecov Report

❌ Patch coverage is 77.77778% with 4 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
.../kotlin/dev/kdriver/core/browser/DefaultBrowser.kt	0.00%	0 Missing and 2 partials ⚠️
...ain/kotlin/dev/kdriver/core/browser/ProcessExit.kt	92.30%	0 Missing and 1 partial ⚠️
...ain/kotlin/dev/kdriver/core/browser/Process.jvm.kt	66.66%	0 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!