feat(airflow): Add semantic inference verification for MaxText E2E DAGs

[chiajunglien]

Description

Issue:

Currently, our Airflow E2E DAGs (e.g., maxtext_e2e_tpu_post_training.py) rely solely on exit 0 to validate job success. However, during recent model testing (e.g., Llama3.1-70B and Gemma3-4B), we discovered "silent failures" where the inference process completes successfully but generates invalid text, nonsensical repetitions, or produces completely empty output.

Changes:

1. Semantic Inference Validation Task: Implemented a decoupled validate_semantic_inference_output
   task that securely queries logs from Cloud Logging for the workload-container.
2. Robust Quality Checks: Added regex parsing to isolate the final Generated text: output block,
   detect non-UTF-8 corruptions (\ufffd), and ensure expected semantic keywords (e.g., London
   landmarks) are present.
3. Directory Restructuring: Moved the utility functions to dags/multipod/util/validation_util.py to
   maintain a clean directory structure.

Tests

https://06ba93284e31466eb62067a2f46710a7-dot-us-east1.composer.googleusercontent.com/dags/test_inference_validation/grid?search=test_inference_validation

Checklist

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

• I have performed a self-review of my code.
• I have necessary comments in my code, particularly in hard-to-understand areas.
• I have run one-shot tests and provided workload links above if applicable.
• I have made or will make corresponding changes to the doc if needed.