[Bug Fix] Janus-Pro runnable with MS2.6/2.7

[zhtmike]

What does this PR do?

Make examples/janus to be runnable with MS2.6/2.7

Fixes # (issue)

• The updated Llama model in Transformers requires setting _support_cache_class=False to be compatible with graph mode. We manually added this attribute in inference.py and generation_inference.py.
• The legacy enable_compile_cache=True setting caused some unknown errors, so we removed it.
• There were some code formatting errors in CI due to an internal .toml file, which we resolved by removing it.
• For text-only training, we fixed the broadcast error related to the image mask.
• For training with mixed tasks, we discontinued support for graph mode with dynamic inputs because: 1) PyNative mode offers better performance; 2) The graph mode interface for dynamic inputs is broken in MS2.7.

Adds # (feature)

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline?
• Did you make sure to update the documentation with your changes? E.g. record bug fixes or new features in What's New. Here are the
  documentation guidelines
• Did you build and run the code without any errors?
• Did you report the running environment (NPU type/MS version) and performance in the doc? (better record it for data loading, model inference, or training tasks)
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

@xxx

[zhtmike]

/gemini review

[gemini-code-assist]

Code Review

This pull request makes the Janus examples compatible with MindSpore 2.6/2.7. The changes include updating environment and performance documentation, applying necessary compatibility fixes for graph mode in newer MindSpore versions, resolving a broadcast error during training, and cleaning up script arguments and configurations. The changes are well-justified and align with the PR's objective. I've added one comment regarding a minor naming inconsistency for future code maintainability improvement.