feat(voice): add native macOS say TTS engine option

[camerondgray]

🗣️ feat(voice): add native macOS say TTS engine option

Summary

Adds an opt-in voice_engine setting to the Pulse voice module so PAI can speak through the native macOS say binary instead of the ElevenLabs cloud API. Setting voice_engine = "say" makes all voice notifications free, fully offline, and key-free. Default behavior is unchanged.

🎯 Motivation and Context

Problem:

• ElevenLabs TTS is metered — every notification costs credits, and hitting the quota cap silently breaks voice (a 401 with no fallback).
• Privacy: every notification's text is POSTed to api.elevenlabs.io. For a personal infrastructure tool, sending all assistant speech to a third party is undesirable for many users.
• It's currently the only engine — there is no offline/local option.

Solution:
Introduce a voice_engine setting ("elevenlabs" | "say"). The "say" engine routes TTS through /usr/bin/say — no API key, no network, nothing leaves the machine. ElevenLabs stays the default, so existing installs are unaffected.

📋 Changes

Single file: Releases/v5.0.0/.claude/PAI/PULSE/VoiceServer/voice.ts

• VoiceConfig gains voice_engine?: "elevenlabs" | "say" and default_say_voice?.
• VoiceEntry gains sayVoice? so each daidentity.voices.<name> can map to a macOS voice.
• New speakWithSay() + sayRateFromSpeed(); playAudio() refactored to share a playAudioFile() helper so the say path keeps the existing per-voice volume control.
• sendNotification() branches on the engine. The ElevenLabs path and the /notify HTTP contract are unchanged, so existing callers need no edits.
• startVoice() resolves the engine (config → PAI_VOICE_ENGINE env → default elevenlabs), warns on an unrecognized value, and only requires an API key for the ElevenLabs engine.
• voiceHealth() reports the active engine.

⚙️ Usage

# PULSE.toml
[voice]
enabled = true
voice_engine = "say"            # "elevenlabs" (default) | "say"
default_say_voice = "Samantha"  # optional; omit to use the system voice

Or via environment: PAI_VOICE_ENGINE=say (and optionally PAI_SAY_VOICE=Samantha).
Optional per-voice mapping in settings.json:

"daidentity": { "voices": { "main": { "voiceId": "…", "sayVoice": "Samantha" } } }

Run say -v '?' to list installed voices; higher-quality voices can be added in System Settings → Accessibility → Spoken Content.

✅ Benefits

• Free — no ElevenLabs usage/credits.
• Private — notification text never leaves the machine.
• Resilient — no API key or quota dependency, and no network round-trip (lower latency).
• Non-breaking — default stays ElevenLabs; existing configs and all /notify callers are unaffected.

🧪 How Has This Been Tested?

• voice_engine="say" speaks via /notify → handleVoiceRequest with no API key set (system voice and a named "Samantha" voice both verified audibly)
• Default (unset) still resolves to elevenlabs — confirmed via voiceHealth()
• PAI_VOICE_ENGINE=say environment selection works
• Unrecognized engine value logs a warning and falls back to elevenlabs
• Per-voice volume preserved (say → AIFF → afplay -v); pronunciation preprocessing applied on both engines

📊 Types of Changes

• New feature (non-breaking change which adds functionality)
• Bug fix (non-breaking change which fixes an issue)
• Breaking change (fix or feature that would cause existing functionality to change)
• Documentation update

✅ Checklist

• My code follows the PAI code style
• I have tested this change thoroughly
• This change is backward compatible
• No shell interpolation — say is invoked via spawn() with an argument array, and -- terminates option parsing so message text can never be treated as flags (consistent with the execSync→execFileSync hardening in security: replace execSync with execFileSync in tab-setter.ts #1046)

🖥️ Platform

macOS only (say is a macOS binary). On non-macOS hosts the engine should be left at the elevenlabs default; selecting say there fails gracefully through the existing voice-error path.