Move flashinfer-python to optional extra vllm[flashinfer]

[mgoin]

Adding flashinfer-python by default to the CUDA requirements introduced a new hard requirement for vLLM to have nvcc installed, which we don't want to enforce for all users. See issue #21960

For now we will move the dependency as an extras i.e. uv pip install vllm[flashinfer]

[gemini-code-assist]

Code Review

This pull request aims to make flashinfer-python an optional dependency by moving it to an extra vllm[flashinfer]. The changes in requirements/cuda.txt and setup.py correctly reflect this. However, there's a significant inconsistency with the Dockerfile, which still installs flashinfer from source unconditionally. This should be addressed to ensure the dependency is truly optional for all users, including those using the official Docker images.

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀