Fix crash in CPU-only environments by skipping torch.cuda.synchronize()

[rayfiyo]

Checklist

• Read CONTRIBUTING.md, and accept the CLA by including the provided snippet. We will not accept PR without this.
• Run pre-commit hook.
• If you changed Rust code, run cargo check, cargo clippy, cargo test.

PR Description

Overview

This PR fixes a crash that occurs when running on CPU-only environments
(e.g., --device cpu or no NVIDIA driver).
Previously, torch.cuda.synchronize() was always called in warmup(),
leading to an AssertionError on CPU builds.

Changes

• In server.py → warmup(), the torch.cuda.synchronize() call is now
  wrapped with a condition checking if --device contains "cuda".

Testing

• Confirmed no crash on a CPU-only setup (pip install torch --index-url https://download.pytorch.org/whl/cpu).
• GPU usage remains unaffected.

Actual command

I tested on Docker's debian:12.5-slim

pip install --upgrade pip --break-system-packages
pip install torch torchvision torchaudio --break-system-packages --index-url https://download.pytorch.org/whl/cpu
pip install moshi --break-system-packages
python3 -m moshi.server --device cpu

CLA

I, rayfiyo, confirm that I have read and understood the terms of the CLA of Kyutai-labs,
as outlined in the repository's CONTRIBUTING.md, and I agree to be bound by these terms.