TTS provider: ElevenLabs v3 (premium)

[thefourcraft]

Part of the Read Aloud (TTS) epic. Premium expressive tier (GA Feb 2026, 70+ langs, cloning).

• Reuse existing elevenlabs key.
• POST /v1/text-to-speech/{voice_id} + WS /stream-input; output_format=pcm_24000.
• Fetch user voices via /v1/voices; mark premium.
• Note: v3 is high-latency (not realtime) — acceptable for read-aloud; optionally expose Flash v2.5 as a 'fast' model id.

File: TextToSpeech/Providers/ElevenLabsTTSProvider.swift.