LocalAI 4.12.0 Streams Qwen TTS Early

This is meaningful latency plumbing for self-hosted voice agents, but not a benchmark proving faster end-to-end performance.

• –Native Qwen streaming uses the `TTSLive` path, with cancellation and backpressure propagated through the pipeline.
• –Other TTS engines require opt-in clause chunking and may not match Qwen’s incremental behavior.
• –The presets do not bundle model weights, so hardware requirements, setup complexity, and model licenses remain important.
• –Deployers should evaluate v4.12.1 instead, which followed with a security fix for external URL handling.