[Bugfix] Fix request disconnect check when BaseHTTPMiddleware is present

[tomeras91]

This PR fixes #10087, by using a different method to check if a request has disconnected (instead of starlette's Request.is_disconnected(), which doesn't work as expected in case a BaseHTTPMiddleware is added to the server - see here - encode/starlette#2094).

The PR uses the method suggested in encode/starlette#2094 (comment)

Added a test that fails on main, and is fixed with this PR

Benchmark

Model: Qwen/Qwen2.5-1.5B-Instruct
Hardware: single H100
Serve run command: vllm serve Qwen/Qwen2.5-1.5B-Instruct

Endpoint: v1/completions
(benchmark run command: python3 benchmark_serving.py --model Qwen/Qwen2.5-1.5B-Instruct --dataset-path ShareGPT_V3_unfiltered_cleaned_split.json)

branch	throughput (requests/s)	mean TTFT (msec)	mean TPOT (msec)	Total generated tokens
main (commit 82c73fd)	52.12	6117.88	24.92	189868
pr (commit 0ba0ff8)	52.52	6064.14	24.20	190241

Conclusion: The fix doesn't hurt performance at all. I guess this is because request.recieve() used here is generally non-blocking

[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.

🚀

[tomeras91]

CC @njhill

[DarkLight1337]

As per offline discussion, we have decided to fix it via #11190 instead.