[Minor][Spec Decode] Remove compiled_softmax

[WoosukKwon]

It turns out that torch-compiled softmax does not provide noticeable speedups:

$ python bench/bench_softmax.py 
VOCAB_SIZE: 131072, NUM_REQS: 16
PyTorch softmax: 0.034 ms
Compiled softmax: 0.040 ms

VOCAB_SIZE: 131072, NUM_REQS: 32
PyTorch softmax: 0.037 ms
Compiled softmax: 0.041 ms

VOCAB_SIZE: 131072, NUM_REQS: 64
PyTorch softmax: 0.061 ms
Compiled softmax: 0.082 ms

Therefore, this PR removes it.

[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.

🚀

[LiuXiaoxuanPKU]

LGTM!