v0.5.1

New Features

• Parquet I/O Across Engines: Added read_parquet/write_parquet support for Local/MultiProcess/Ray/Spark DataFrames and dataset loader.
• LLM-Based Sample Filtering: Added LLMJudgeFilter for guidance-driven keep/drop decisions with metrics output.

What's Changed

• refactor: Improve JSON extraction from markdown by using regex
• feat: Add AGENTS.md contributor guide outlining and operator generator skill
• feat: add ChatContextRoundTrimmer to trim chat context by rounds
• feat: add dynamic max_rounds support to ChatMultiTurnSynthesizer
• feat: add persona-based chat synthesis example
• feat: implement LLM-based sample filter
• feat: add MetadataClearer to clear metadata
• test: add pytest suite and initial unit tests
• test: add more unit tests
• test: add unit tests for dataframe & dataset & provider
• test: align test filenames with source modules
• docs: update operator generate SKILL.md to include unit testing guidelines
• fix: correct ray dataframe union and add spark/ray tests
• feat: add parquet support across engines