Reap leftover EVE VM before running eden tests

[europaul]

Problem

Self-hosted runners were getting wedged: several kratos-runner-eden-* runners had a leftover qemu-system-x86_64 (an EVE VM) holding the console port 17777, so every subsequent eden test job on that runner failed at eden start with the port already in use.

Tracing it back, the orphaned VMs all came from the Smoke suite's eve_restart test in jobs that were cancelled mid-test (e.g. a newer PR-Gate run superseding an in-flight one).

Why the existing cleanup doesn't handle it

The Clean up after test step in this composite action (./eden stop etc.) is guarded by if: always():

• always() runs on success and failure, but a composite action's remaining steps are not executed when the job is cancelled — the whole action is torn down mid-run. So on cancellation the VM is never stopped.
• It also can't be reaped after the fact: eden tracks the VM through a context stored under the ephemeral HOME of the cancelled test run, which is gone. A later job's eden stop / make clean finds no context and skips stopping the process. Verified on the stuck runners — running both eden stop and make clean were no-ops and the qemu kept running; only freeing the port directly reaped it.

Fix

Add a pre-test step that frees port 17777 and cleans the workspace before Setup Environment, so a leftover from a previous job can't block the next one. Killing by the well-known EVE console port is surgical — it only targets an orphaned EVE VM, not any unrelated qemu.

- name: Clean up before test
  run: |
    fuser -k 17777/tcp || true
    make clean || true

[europaul]

@rene we should add "stable" label to Eden github repo. for example for PRs like this one

[rene]

@rene we should add "stable" label to Eden github repo. for example for PRs like this one

yes, I've created the label and added to this PR as well.