ci: retry jobs on transient infrastructure failures

Add retry to the default: block so every job automatically retries up to twice on:

• runner_interrupted — runner was restarted or the host was reclaimed mid-job
• runner_external_dependency_failure — runner couldn't reach an external dependency (image registry, apk mirrors, etc.)

Script failures, auth errors, and configuration errors are intentionally excluded, so retries won't mask real problems like a Go version mismatch or a failing test.