Add a batched auto tune script

[karan]

Purpose

Make it easy to run multiple auto tune experiments.

Test Plan

Tested locally against the real auto_tune.sh script.

Test Result

Successfully ran 6 back to back auto_tune.sh experiments.

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

[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 ask your reviewers to trigger select CI tests on top of fastcheck CI.

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.

If you have any questions, please reach out to us on Slack at https://slack.vllm.ai.

🚀

[gemini-code-assist]

Code Review

This pull request introduces a batch auto-tuning script, batch_auto_tune.sh, which is a great addition for running multiple experiments. The accompanying documentation update in the README is clear and comprehensive.

My review focuses on improving the robustness and correctness of the new bash script. I've identified a critical issue with how the script saves progress, which could lead to data loss, and a couple of high-severity issues related to path resolution and input validation that could cause the script to fail unexpectedly. The suggested changes will make the script more reliable and easier to use.

[yaochengji]

LGTM, thanks for the improvement!