[Misc] Further clean up some redundant config definitions

[Isotr0py]

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.

Purpose

Related issue: #18850, also cc @princepride

• Some HF config like VisionEncoderDecoderConfig used by ByteDance/Dolphin enforces has_no_defaults_at_init=True, so we can't tie it with trust_remote_code.
• This PR untie has_no_defaults_at_init from trust_remote_code and add method to override kwargs when initialize AutoConfig.
• With these changes, we can further clean up some unnecessary config porting.

Test Plan

Test Result

(Optional) Documentation Update

[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.

🚀

[DarkLight1337]

Thanks for further simplifying this!

[gemini-code-assist]

Code Review

This pull request refactors the model configuration loading by removing several custom configuration classes and introducing a more generic mechanism to override AutoConfig keyword arguments. This is a positive change that simplifies the codebase. However, I have identified two potential issues in vllm/transformers_utils/config.py that are either introduced or exposed by these changes. One is a semantic change in is_encoder_decoder that removes its recursive behavior, which could be problematic for models with nested configurations. The other is a critical issue in _maybe_remap_hf_config_attrs where a sub-configuration is handled as a dictionary instead of a PretrainedConfig object, which will likely cause runtime errors.