[V1] Enable Inductor when using piecewise CUDA graphs

[WoosukKwon]

Unfortunately, after the PR #10228 piecewise CUDA graph generates gibberish outputs. For some reason, we have to turn on the inductors and disable the custom CUDA kernels when using piecewise CUDA graphs.

Note that the following error message appears when using the CUDA graphs in V1 (although it functionally works):

[rank0]:W1112 00:26:33.338000 39792 site-packages/torch/_inductor/codecache.py:604] [0/0] Can't pickle
[rank0]:W1112 00:26:33.338000 39792 site-packages/torch/_inductor/codecache.py:604] [0/0] Traceback (most recent call last):
[rank0]:W1112 00:26:33.338000 39792 site-packages/torch/_inductor/codecache.py:604] [0/0]   File "/woosuk/miniconda3/envs/vllm/lib/python3.10/site-packages/torch/_inductor/codecache.py", line 600, in dumps

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.
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 do one of these:

• Add ready label to the PR
• Enable auto-merge.

🚀