Fix race condition in IsRunningStartupCheckStrategy with stale cached container state (#11860)

[vpelikh]

Problem

In CI environments (Kubernetes, Jenkins), PostgreSQL (and potentially other) containers fail with:

Wait strategy failed. Container is removed

Root cause: IsRunningStartupCheckStrategy had a short-circuit optimization that used container.getContainerInfo().getState() — which returns stale cached state from the preceding port-mapping check. If the container exits/crashes between the port-mapping check and the startup check:

1. Port-mapping check caches containerInfo with state "running"
2. Container exits
3. IsRunningStartupCheckStrategy reads stale "running" state → startup passes incorrectly
4. Wait strategy starts (docker logs --follow) on crashed/removed container → NotFoundException
5. User sees "Container is removed" with no indication of the actual failure

Fix

Don't blindly trust the cached state — verify it with one live Docker inspect before declaring success. The cached state from the preceding port-mapping check is used as a hint (fast path), but is confirmed via a single checkStartupState call. If the live inspect confirms the container is running, we return success immediately. If the live detect shows a different state (stale cache), we fall through to full rate-limited polling. If the live inspect fails/timeout (e.g., Docker unresponsive on slow CI), we gracefully fall back to trusting the cached state as the best available information.

Additional improvements:

• Include containerId in the "Container is removed" error message for better diagnostics
• Wrap getLogs() and stop() in individual try-catch blocks during cleanup to prevent cascading failures (e.g., NotFoundException when the container is already removed) from suppressing the original ContainerLaunchException
• Re-enable testCommandQuickExitFailure (was @Disabled due to flakiness from the cached-state race)
• Add testQuickExitWithDifferentExitCode to validate any non-zero exit code is detected

Design Rationale

The hybrid approach was chosen over two alternatives:

1. Remove cached-state shortcut entirely — The simplest fix for [Bug]: PostgreSQL container intermittently fails to start with "Wait strategy failed. Container is removed" (TimeoutException) in CI environment #11860, mine original implementation. It caused consistent CI failures on CircleCI where docker inspect hangs, making every startup check timeout.
2. Keep the cached-state shortcut as-is — Would pass CI but doesn't fix the stale-state bug.
3. Hybrid (implemented) — Cached state as hint, verify with one live Docker inspect, catch timeout → trust cache. This fixes the stale-state bug when Docker is responsive, while remaining resilient to Docker inspect timeouts observed on CircleCI's machine executor.

Changes

• IsRunningStartupCheckStrategy.java: Replaced blind cached-state shortcut with verify-then-trust hybrid — uses cached state as hint but confirms with one live Docker inspect before returning success
• GenericContainer.java: Better error message; NotFoundException-safe cleanup
• IsRunningStartupCheckStrategyTest.java: Re-enabled flaky test; added test for non-zero exit code