Onboarding Gemma 4 in V1

### Reminder - [x] I have read the above rules and searched the existing issues. ### Description Thanks for maintaining this amazing project! Right now, it seems that Gemma 4 support is only done in V0. To fully realize the 256K context length, we need to enable Context Parallelism via deepspeed-ulysses, which is only available in V1. To that end, we need to register Gemma4 for 1) reasoning template class and 2) tool call rendering. Please feel free to let me know if there is already some ongoing efforts in this direction, or missing anything. ### Pull Request No, but happy to contribute