Fix connection lifecycle: disconnect propagation, connect dedup, client cleanup

[nathanfallet]

Summary

Three connection-ownership issues, sharing the lifecycle root cause. Builds on the WebSocketTransport seam + pendingRequests registry from #73.

ISSUE-3 — disconnect/close hung in-flight commands

When the WebSocket dropped or close() was called, the receive loop just ended; any callCommand parked on its response deferred waited forever (the loop only printStackTrace'd). Now failPendingRequests(...) drains the registry and completes every waiter exceptionally with a new ConnectionClosedException:

• on normal completion of incoming() (socket closed),
• on error in the receive loop (now logger.error, not printStackTrace),
• and in close().

Callers observe a failure instead of hanging.

ISSUE-4 — unsynchronized connect() opened duplicate sessions

Concurrent first commands could each pass the isActive check and open a second session, leaking the extra socket + listener. connect() is now guarded by a double-checked connectMutex; the prepareExpert/prepareHeadless block by a separate prepareMutex (distinct mutex avoids re-entrancy — prepare* issue ONE_SHOT commands that skip that block).

ISSUE-9 — HttpClients never closed

KtorWebSocketTransport.close() now closes its engine-backed HttpClient; DefaultBrowser.stop() closes the HTTP-API client. createBrowser/stop cycles no longer leak threads/connection pools.

Verification (red → green)

New ConnectionLifecycleTest (fake transport, UnconfinedTestDispatcher) — all three fail against the unfixed code and pass after:

• callCommand fails with ConnectionClosedException on disconnect (RED: hang → timeout).
• callCommand fails with ConnectionClosedException on close (RED: hang → timeout).
• two concurrent first commands open the transport exactly once (RED: count == 2).

ISSUE-9 is resource cleanup — not deterministically unit-testable without injecting clients, so verified by inspection plus the real-browser createBrowser/stop suite.

Full :core:jvmTest (real Chrome) + :opentelemetry:jvmTest pass; macOS native tests pass; Linux/MinGW compile.

Test plan

• ./gradlew :core:jvmTest --tests "*.ConnectionLifecycleTest" — red on unfixed code, green after
• ./gradlew :core:jvmTest (real Chrome) — all pass
• ./gradlew :opentelemetry:jvmTest — all pass
• ./gradlew :core:macosArm64Test + Linux/MinGW compile — pass

[codecov]

Codecov Report

❌ Patch coverage is 80.00000% with 5 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...n/dev/kdriver/core/connection/DefaultConnection.kt	77.77%	2 Missing and 2 partials ⚠️
.../kotlin/dev/kdriver/core/browser/DefaultBrowser.kt	66.66%	0 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!