feat: say how the browser died, not just that it is gone

[nathanfallet]

The problem

Since #83 a failed start reports alive=false: we know Chrome is gone, not how it went. The two possible deaths call for opposite investigations.

In production on 27/08, on the VMs still failing, the message is invariably:

Browser never opened its debug port on 127.0.0.1:53997 after 30500ms
(pid=8932, alive=false). Last connection error: ConnectException: Connection refused.
Browser stderr: <none>

Chrome starts, dies within 30s, and writes nothing. Did it crash, or did it stop on purpose? That decides where to look next, and today we cannot tell.

The change

Process.exitCodeOrNull() on all four targets:

• jvm — exitValue(), which throws IllegalThreadStateException while the process lives: that is exactly the "still running" case, reported as null.
• mingw — GetExitCodeProcess, the same call isAlive() already uses; null while it reports STILL_ACTIVE.
• posix — null, deliberately: liveness there is kill(pid, 0), which does not carry a status, and nothing reaps the child to collect one.
• js — unsupported, like the other process functions.

The status is reported verbatim, with no interpretation. There is a plausible reading of an exit 0 on Windows — Chrome hands its command line to an instance that already owns the same --user-data-dir and quits without opening a port — but that is a hypothesis, not something this PR asserts. Reading the status is the whole point: we do not know yet which statuses this failure produces, and putting a guess in every log line would defeat the instrumentation.

Scope

Message building moves to StartFailure.kt as two pure functions, processFate and browserStartFailureMessage, so the wording is covered by tests on any OS — the facts they report come from platform calls no CI can exercise.

Nothing changes on the happy path: both are only called in the failure branch.

Tests

Endpoint and pid always present; exit status reported verbatim for both 0 and non-zero; still-running said plainly; an unavailable status must not be reported as an exit code; stderr blank vs populated; last connection error.

All five targets compile, :core:jvmTest green, detekt 317 findings, same as main.

[codecov]

Codecov Report

❌ Patch coverage is 35.71429% with 9 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
.../kotlin/dev/kdriver/core/browser/DefaultBrowser.kt	0.00%	6 Missing ⚠️
...in/kotlin/dev/kdriver/core/browser/StartFailure.kt	71.42%	0 Missing and 2 partials ⚠️
...ain/kotlin/dev/kdriver/core/browser/Process.jvm.kt	0.00%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!