substrate-step-test: after a FAIL the lab holds a Session that cannot be deleted #567

Problem

After a substrate-step-test FAIL, stepping the lab back to the chart's Substrate left the drain session's Actor NotFound while kagent still held its RESUME operation. The lab then holds a Session that cannot be deleted (kagent never completes the operation against a missing Actor) until a kagent carrying the resume-retry fix (kagent-upstream#417) is installed.

Proposed solution

After a FAIL, substrate-step-test removes the sessions it created (or marks the lab dirty and says which sessions block a reset) before stepping back; the lab's reset path force-clears a Session whose Actor is NotFound.

Acceptance criteria

• After a FAIL and the step back, no Session blocks deleting the test's namespace or a fresh run.
• If one does, the command names it and the way to clear it.

This issue was written by an agent.