v0.5.1

## Fixes

- The CLI now selects GGML when CUDA is unavailable, so `faster-qwen3-tts custom ...` works on a supported Mac without `--backend ggml`. CUDA hosts still default to the Torch backend; explicit backend selection remains available.
- A standard install on native Apple Silicon now includes `qwentts-cpp-python>=0.4.1` and its Metal runtime. Other platforms still use the optional `ggml` extra.

Upgrade with `uv pip install -U faster-qwen3-tts`. The Metal wheel supports macOS 14+ on ARM64 Python; Intel Macs and Rosetta Python are outside its supported platforms.

## Changes

- [Make Apple Silicon GGML CLI work by default](https://github.com/andimarafioti/faster-qwen3-tts/pull/140)
- [Prepare release 0.5.1](https://github.com/andimarafioti/faster-qwen3-tts/pull/141)

[Full changelog](https://github.com/andimarafioti/faster-qwen3-tts/compare/v0.5.0...v0.5.1)