Stale ssh Host alias from a previous cloud VM silently points new runs at a dead IP

Symptom: a freshly launched cloud VM's staging step and watcher kept timing out, although the instance was healthy.

Cause: the ssh config still held a Host entry with the same alias from the previous run, pointing at the old, terminated VM's IP. A new entry for the same alias had been appended, but ssh uses the FIRST matching Host block for each option. So HostName resolved to the dead machine.

Fixes: - Give each run a unique alias (e.g. include the run id), or delete the old block before adding a new one. Don't just append. - Check with ssh -G <alias> | grep -i '^hostname' right after writing the config, before staging starts. - After fixing mid-run, confirm the timeouts didn't trip watcher limits, terminate anything, or mark a partial pull as final. Also re-check that the 'pull before cap' margin still holds after the late start.