0.2.5

## What's Changed
Everything! since this is the first release on github. But I think since 0.2.4:

* fix: require torch>=2.5.1 for CUDA graph reliability by @andimarafioti in https://github.com/andimarafioti/faster-qwen3-tts/pull/53
* Fix sample-rate mismatch causing low/slow audio in OpenAI server by @andimarafioti in https://github.com/andimarafioti/faster-qwen3-tts/pull/54
* feat: add instruct parameter to generate_voice_clone by @andimarafioti in https://github.com/andimarafioti/faster-qwen3-tts/pull/40
* Fix generate_with_embedding speech tokenizer access by @andimarafioti in https://github.com/andimarafioti/faster-qwen3-tts/pull/61
* feat: accept precomputed voice_clone_prompt in public API by @andimarafioti in https://github.com/andimarafioti/faster-qwen3-tts/pull/56
* fix(base): align voice clone defaults with upstream by @andimarafioti in https://github.com/andimarafioti/faster-qwen3-tts/pull/69
* Add StreamPlayer helper and clarify streaming playback docs by @andimarafioti in https://github.com/andimarafioti/faster-qwen3-tts/pull/78
* docs(setup): update Blackwell guidance to cu128 by @andimarafioti in https://github.com/andimarafioti/faster-qwen3-tts/pull/67
* Add demo text-feed toggle for non_streaming_mode by @andimarafioti in https://github.com/andimarafioti/faster-qwen3-tts/pull/90
* Fix PyPI license metadata by @andimarafioti in https://github.com/andimarafioti/faster-qwen3-tts/pull/91


**Full Changelog**: https://github.com/andimarafioti/faster-qwen3-tts/commits/0.2.5