What does this PR do?

Follow-up to #13663 (per @asomoza's review there): replace the AutoModel / AutoModelForCausalLM references for text_encoder and pe with the concrete Mistral3Model / Ministral3ForCausalLM so the loaded checkpoint matches the declared types.

To avoid breaking installs with older transformers (Ministral3ForCausalLM is only in 5.0+, Mistral3Model is in 4.50+):

• Standard: TYPE_CHECKING + string annotations on __init__ so the imports never run at module load. _get_signature_types uses get_type_hints and falls back gracefully when a name can't be resolved.
• Modular encoders.py: try/except import that aliases the concrete class when available and falls back to AutoModel / AutoModelForCausalLM otherwise. The ComponentSpec then references the alias.

So users on recent transformers get the silenced warning; users on older transformers keep working as before.

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 read our philosophy doc (important for complex PRs)?
• Was this discussed/approved via a GitHub issue? Address ernie-image review findings #13577 #13663
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

Who can review?

@yiyixuxu @asomoza