Add TIPSv2 Scenic demo

[bingyic]

Summary

Add a unified Scenic (JAX/Flax) demo notebook supporting both TIPSv1 and TIPSv2 models.

Supersedes #10 with a cleaner implementation that leverages the configurable posembs parameter from #11.

Key Changes

Notebook (scenic/notebooks/TIPS_Demo.ipynb)

• Unified demo supporting TIPS and TIPS_V2 model types
• Model config (tips_model_config.py) handles V1 vs V2 positional embedding sizes automatically
• No monkey-patching: TIPSv2's (32, 32) positional embeddings are passed cleanly via:

  model_vision = tips_model.VisionEncoder(
      variant=model_config.variant,
      posembs=model_config.positional_embedding.shape,  # (16,16) for V1, (32,32) for V2
  )

Checkpoint Downloads

• Added TIPSv2 checkpoint URLs for both PyTorch and Scenic formats
• Updated download_checkpoints.sh scripts

Model Config (tips_model_config.py)

• Added TIPSv2 variant mappings (B/14, L/14, So400m/14, g/14)
• Config automatically sets correct positional_embedding.shape per model version

Co-authored-by: washingtonsk8

[bingyic]

I've force-pushed the correct commit (e2c7264) to the scenic_demo branch, replacing the accidental merge commit. The PR is now ready for review.