[Misc] Benchmarks for audio models

[NickLucche]

Implements feature requested in #16354.

Test with:

# server
vllm serve openai/whisper-large-v3-turbo

# client
python3 benchmarks/benchmark_serving.py \       
    --backend openai-audio \
    --dataset-name hf \
    --dataset-path edinburghcstr/ami --hf-subset ihm \
    --model openai/whisper-large-v3-turbo \
    --num-prompts 1000 \
    --endpoint /v1/audio/transcriptions \
    --save-result \
    2>&1 | tee benchmark_whisper.txt

It's still a draft because I want to sweep through the datasets first, but reviews are welcome!
cc @DarkLight1337

FIX #16354

[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.

🚀

[NickLucche]

PS: these datasets need to be granted access manually on hf. Posting here in case we decide to run performance checks on them:
https://huggingface.co/datasets/speechcolab/gigaspeech
https://huggingface.co/datasets/kensho/spgispeech

[DarkLight1337]

Can you merge from main to try to fix the CI errors?