Fix Gemma-3 4B Multimodal SFT test script

[RexBearIU]

Description

This PR fixes multiple bugs and configuration issues in Step 4 of the Gemma-3 4B Multimodal SFT end-to-end test script:

1. Fixes Syntax Typo: Corrects the typo run_name=${run_id}} (with a trailing curly bracket) to run_name=${run_id}.
2. Adds Missing Tokenizer Configuration: Adds tokenizer_type=huggingface to Step 4. Without this, MaxText defaults to a SentencePiece tokenizer and attempts to load a local .model file at the Hugging Face repo path google/gemma-3-4b-it, resulting in the following error:

   tensorflow.python.framework.errors_impl.NotFoundError: google/gemma-3-4b-it; No such file or directory
   

Description of the issue we found

Without the tokenizer_type=huggingface flag, Step 4 of the end-to-end Gemma-3 SFT test crashes upon parameter loading when it tries to find a local SentencePiece model path under google/gemma-3-4b-it instead of fetching the proper Hugging Face config.

Tests

We tested this change end-to-end on a v6e TPU VM:

1. Re-ran the end-to-end script tests/end_to_end/tpu/gemma3/4b/test_gemma3_multimodal_sft.sh using a unique run ID.
2. Verified that Step 2 successfully executed inference.
3. Verified that Step 3 completed all 5 SFT training steps and successfully saved the checkpoints to GCS.
4. Verified that Step 4 successfully completed loading parameters from the newly trained SFT checkpoint on GCS, initialized the Hugging Face tokenizer, and correctly decoded/described the image:

   Input `<start_of_turn>user
   Describe image <img><end_of_turn>
   <start_of_turn>model
   ` -> `Seattle skyline`
   

Checklist

Before submitting this PR, please make sure (put X in square brackets):

• I have performed a self-review of my code. For an optional AI review, add the gemini-review label.
• I have necessary comments in my code, particularly in hard-to-understand areas.
• I have run end-to-end tests tests and provided workload links above if applicable.
• I have made or will make corresponding changes to the doc if needed, including adding new documentation pages to the relevant Table of Contents (toctree directive) as explained in our documentation.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!