fix: Replace a test container whose database does not become reachable, and remove every container that failed to start.

[goloroden]

A local make qa failed in verifiesTheHashOfAnEventWhoseDataTheServerEscapes, because its container did not start:

Timed out waiting for URL to be accessible (http://localhost:63085/api/v1/ping should return HTTP 200)

Docker Desktop sometimes does not make the port it maps a container to reachable from the host. The container keeps running, but the database in it can never be reached. start() then failed after 10 seconds, and left the container behind. The Go SDK fixed the same in #188.

What changes:

• start() replaces such a container with a new one, up to three times.
• If the third container fails, too, start() throws an IllegalStateException that says so, with the last failure as its cause.
• Every other failure, e.g. an image that does not exist, or a database that exits right away, is thrown at once, as it is.
• Either way, the container that failed to start is removed, so that it is not left behind.
• Whether a failure is worth another attempt depends on whether the container still runs, which is what Go checks, too.
• The public API is unchanged. Tests can replace how a container is started through a package-private field, as Go does with startContainer.

Tests:

• New: replacing a container twice and then starting, giving up after three attempts, and throwing another failure at once. Each test also checks that no failed container is left in Docker.
• Two mutations are caught: one attempt instead of three, and not removing a failed container.
• make qa is green, and coverage stays at 100 %.

This will conflict with #2, which documents Container, in a few lines around the constructor and start(). Whichever merges second, I rebase.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Hc2MPSLu7HDHTm8HiBsrDz