Validate Portable Hermes lifecycle and click-to-first-prompt (PR #11578 + performance PR)

Parent epic: #11573

## Outcome

Produce exact-commit Portable evidence for lifecycle reliability, Play click-to-first-prompt latency, and a useful first response.

## Current local evidence on `30ef9ec7a`

- Two consecutive starts: 6.98s and 6.93s.
- Two separate authenticated probes: 1.72s and 1.73s.
- Cold Hermes TUI displayed its input prompt at approximately 1s.
- A real prompt returned the requested `READY`; generation displayed 11s.
- The earlier 22.9s arithmetic estimate is not a Play click-to-prompt measurement because it sums work that overlaps Portable seat setup.
- The current 16–20s expectation remains a projection, not acceptance evidence.

## Pull-request relationship

This task owns exact-commit lifecycle and timing acceptance for the fixes delivered by merged PR #11578 and the planned Portable performance PR tracked by #11574.

Performance requirements from #11570, #11571, and #11572 are consolidated in #11574; this validation task remains separate.

## Required evidence

- [ ] Run a real Portable Play cycle on exact commit [`30ef9ec7a`](https://github.com/NVIDIA/NemoClaw/commit/30ef9ec7a198270b9524ddaa8c4d0aef9b0da918).
- [ ] Record Click Play → Rig Ready, prewarm start/end, autostart launch, first usable prompt, and first useful response separately.
- [ ] Run at least three stop/start/probe/launch cycles and report variance.
- [ ] Run one fresh onboarding and one resume/restart path without manual state repair.
- [ ] Confirm no HOME-fence, lock-root, provider-version, network-reuse, OpenShell-state, or forward-adoption regression.
- [ ] Confirm ports `18789` and `8642` are adopted only when exact identity matches.
- [ ] Confirm a real response does not reproduce the earlier apparent 60s generation hang.

## Completion

- [ ] Link the final trace and exact accepted commit.
- [ ] Record whether the 20s target passed, failed, or received an explicit disposition.