feat(voice): fall back to macOS say when ElevenLabs unavailable

[blu3dot]

Summary

When ELEVENLABS_API_KEY is missing or the ElevenLabs API call fails, VoiceServer now falls back to the macOS native say command instead of silently skipping voice output. This gives new users immediate voice feedback without requiring ElevenLabs setup.

Problem

VoiceServer silently skips all TTS when ELEVENLABS_API_KEY is missing. New users who haven't configured ElevenLabs get zero voice feedback with no indication why — the server returns HTTP 200, making the failure invisible.

Solution

After the ElevenLabs TTS path is skipped (no API key) or fails (API error), fall back to /usr/bin/say:

• Reuses existing applyPronunciations() so custom pronunciation rules apply to native voice too
• Reuses existing spawnSafe() helper (array-form spawn, no shell injection)
• Only triggers when ElevenLabs path didn't play (no double-speak)
• Fails gracefully — logs error, doesn't crash server
• Uses error: unknown with instanceof type guard (also upgrades the adjacent ElevenLabs catch block)

macOS-only — say is present on all macOS versions. Linux equivalent is out of scope for this PR (see #855, #872 for cross-platform audio work).

Test plan

• Start server with no ELEVENLABS_API_KEY → send notification → hear macOS say voice
• Set invalid ELEVENLABS_API_KEY → send notification → ElevenLabs fails → fallback to say
• Both paths fail (e.g., Linux) → notification still displays, no voice, error logged
• Custom pronunciations from pronunciations.json apply to say output

Post-Deploy Monitoring & Validation

No additional operational monitoring required: additive fallback path, existing behavior unchanged when ElevenLabs key is configured.

---

🤖 Generated with Claude Code

[blu3dot]

Closing to test locally before resubmitting.