Remove V0 Encoder-Decoder Support

[WoosukKwon]

Remove V0 encoder decoder model runner.
Also, this PR deletes the deprecated models such as BART. After this PR, Whisper will be the only encoder-decoder model that are supported by vLLM.

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @WoosukKwon.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[gemini-code-assist]

Code Review

This pull request restores support for Mllama4 and Skywork models. As part of this, it also removes a significant amount of code and tests related to a generic encoder-decoder model runner, effectively dropping support for models like BART, mBART, Donut, Florence2, and the original Mllama. While the changes are extensive, they appear to be mostly consistent with this goal. However, I've identified two critical bugs in an example script that will lead to NameError exceptions at runtime due to variables being used after their definitions were removed.

[mfournioux]

@WoosukKwon your PR is deleting Donut model from offline inference examples. Could you please let me know why Donut model is not longer supported while this PR has been merged recently #23229 ?

[WoosukKwon]

@mfournioux We were planning to discontinue encoder-decoder model support (except Whisper) in the course of V0 deprecation. I think the PR was merged without the lack of this information.

[mfournioux]

@mfournioux We were planning to discontinue encoder-decoder model support (except Whisper) in the course of V0 deprecation. I think the PR was merged without the lack of this information.

Many thanks for your reply, why only Whisper model is kept and others like Donut are not longer supported regarding this PR ? It means Donut model is not longer supported for next vllm version ?