Refactor/vision token utils

[shreed27]

Refactor vision token utils to decouple from tokenizer

Description

Refactors _token_utils.py to accept special_tokens as an argument instead of hardcoding Gemma3Tokenizer.
This addresses the TODO This value should be propagated from the model. and makes the utility completely model-agnostic.

Changes

• gemma/gm/vision/_token_utils.py:
  • Updated add_extra_tokens_for_images to accept special_tokens.
  • Updated remove_mm_logits to accept special_tokens.
  • Removed hardcoded _tokenizer.Gemma3Tokenizer usage.
• gemma/gm/utils/_types.py: Propagate self.config.special_tokens to add_extra_tokens_for_images.
• gemma/gm/nn/_transformer.py: Propagate self.config.input_config.special_tokens to remove_mm_logits.
• gemma/gm/nn/gemma3n/_transformer.py: Propagate self.config.input_config.special_tokens to remove_mm_logits.
• gemma/gm/vision/_token_utils_test.py: Updated tests to use a mock SpecialTokens class.

Verification

• Verified function signatures and basic logic via script (mocking dependencies).
• Unit tests updated to match new signature.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.