Gemma 4 and Qwen TTS support

According to your documentation you support Gemma 4 models https://docs.cactuscompute.com/latest/#supported-llms. I was wondering if we can use a single model instance for generation but also transcription. Instead of instantiating models with the same identifier for different tasks, e.g. a `CattusLLM` instance for completion and a `CactusSTT` for transcription. Ideally, I'd like a single `CactusModel` that supports both `transcribe` and `completion` uses.

Also, I was wondering if support for text to speech would be available through the Qwen 3 tts model family https://qwen.ai/blog?id=qwen3tts-0115

Happy to collaborate!

Start by reviewing the documented supported LLMs and the existing CattusLLM, CactusSTT, and CactusModel interfaces. Determine whether one model instance can expose both completion and transcription, and assess the requested Qwen 3 TTS support; done means the supported model capabilities and unified API behavior are defined and implemented.

Written by the indexing model from the issue text.