[Core] Gate prompt_embeds behind a feature flag

[DarkLight1337]

Since the existence of input_embeds triggers a separate graph compilation, this PR disables prompt_embeds inputs (#15428) by default. This should revert the engine startup time to normal for most users.

[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.

🚀

[njhill]

Thanks @DarkLight1337

[qthequartermasterman]

Asking for clarification. This only affects the availability of prompt embeds if not in eager mode, correct?

If forcing eager mode, no compilation is needed and the prompt embeds are passed to the python implementation of the model regardless of this flag, right?

[DarkLight1337]

For consistency, this flag is required for both eager and graph mode.