Improve flaky waitForRunningContainer test

Why was this MR needed?

!4996 (merged) introduced the waitForRunningContainer test. However, the hard-coded sleeps were too fast on Windows. Use channels and signals to avoid using hard-coded values.

What's the best way to test this MR?

On Windows, macOS, and Linux, run:

go test gitlab.com/gitlab-org/gitlab-runner/executors/kubernetes -run TestWaitForRunningContainer

What are the relevant issue numbers?

Relates to #38016 (closed)