v1.18.0 — Atom TTS prefetch+FIFO, deferred idle TTS, Tailscale router guide

[amariichi]

Summary

Three independently useful changes bundled into the v1.18.0 release.

1. Atom TTS prefetch + FIFO (PLANS_53) — primary feature

• TTS worker: MH_TTS_REMOTE_PREFETCH_MS (default 900) returns play_stop early for browser-only audio targets so the next chunk's synthesis can overlap the current chunk's playback.
• face-app browser audio: same-session chunks queue sequentially instead of replacing the active source.
• AtomS3R firmware: one-pending-WAV FIFO; stop(), PTT, and interrupt flush both the active and queued audio.
• Net effect: long multi-sentence answers on PC browser, mobile browser, and AtomS3R play with much shorter inter-chunk gaps.

2. Bridge generation watermark fix (PLANS_52 × PLANS_53 integration)

Hardware testing surfaced a regression where the AtomS3R cut off mid-utterance and then spoke an idle notification. Root cause: scripts/atoms3r-http-bridge.mjs advanced its generation watermark on every relayed payload, so a deferred-idle (PLANS_52) tts_state at generation N+1 caused the in-flight gen=N utterance's tail chunks to be dropped as "stale". Fix: scope observeGeneration to tts_audio / tts_audio_ref only. Verified end-to-end on hardware.

3. Deferred idle TTS + Tailscale router guide

Already on the branch from earlier work:

• Defer idle TTS notifications until speech drains (PLANS_52)
• Add Tailscale travel router setup guide

Docs and defaults

• README, doc/guides/operator-stack.md (EN+JA), and examples/rmh-voice-mode/README.md now recommend --audio-target browser for PC browser, mobile, and AtomS3R, with a brief note that worker prefetch is disabled in both mode.
• restart-operator-stack-in-place.sh default changed from both to browser to match run-operator-stack.sh's existing default.

Version

Bumped across all six sites: package.json, mcp-server/dist/index.js (SERVER_VERSION), both worker pyproject.toml, both uv.lock (regenerated via uv lock).

Test plan

• node --check scripts/atoms3r-http-bridge.mjs
• .venv-platformio/bin/pio run -d firmware/atoms3r-headroom (per PLANS_53 ExecPlan)
• Full npm test (375 tests, per PLANS_53 ExecPlan)
• Hardware: 3-step live test on AtomS3R confirmed the gap-reduction, the bridge regression, and the post-fix end-to-end success
• Reviewer: spot-check doc/guides/operator-stack.md audio target descriptions render well in both languages

🤖 Generated with Claude Code