[Model] New interface and automatic detection for PP support

[DarkLight1337]

In this PR, I have updated the model registry to import model modules in a separate process to check whether they support PP. This avoids initializing CUDA for the main program. I have also applied this to the check for multimodal models.

With this improvement, I won't have to add almost every model to the hardcoded list for #7168.

While working on this, I have also moved the iteration over architectures from ModelConfig into ModelRegistry. This supersedes #8924.

[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.

🚀

[youkaichao]

ideally we should support pp for all models. I'm waiting for @andoorve to support this.

for multi-modality support, we can use this approach.

[DarkLight1337]

Closing as #7168 will include these changes.