Add multimodal SQL helper UDFs

[JingsongLi]

Summary

Adds Python DataFusion SQL helper UDFs for BLOB-backed multimodal workflows and vector JSON bridging. This builds on the existing video frame extraction support with media metadata, thumbnail generation, and JSON-to-vector conversion helpers.

Changes

• Register Python binding built-ins for media_info, media_thumbnail, vector_from_json, and vector_to_json.
• Add media inspection and PNG thumbnail generation for inline BLOB bytes and resolvable BlobDescriptor inputs.
• Add JSON vector bridge helpers that convert JSON float arrays to Arrow List<Float32> values and back to JSON strings.
• Extend Python DataFusion tests for registration, image/video metadata, thumbnails, invalid inputs, and vector conversion.
• Document the Python-only multimodal helper functions in SQL docs, README, and package description.

Testing

• cargo check -p pypaimon_rust
• cargo fmt --check
• git diff --check
• uv run --no-sync pytest tests/test_datafusion.py

Notes

These helper functions are registered by pypaimon_rust.datafusion.SQLContext; they are not registered by the Rust paimon_datafusion::SQLContext.

Relates to #227.

[leaves12138]

LGTM. Reviewed the multimodal SQL helper UDF implementation and documentation, and verified locally with cargo check plus targeted Python UDF tests.