Deliver v3.0.2

[battmanux]

This pull request introduces several enhancements and fixes across the codebase, focusing on improved Pydantic model handling, type conversion robustness, and new features for meta-prompt tuning and probabilistic output. It also includes a minor version bump and updates to documentation and example notebooks.

Core Library Improvements:

• Enhanced Pydantic support:

  • Added dynamic reconstruction of Pydantic model class definitions as strings with reconstruct_pydantic_class_string_auto, and improved detection of Pydantic models throughout type conversion and schema description functions. [1] [2] [3] [4]
  • Updated type conversion logic to robustly handle Any, Enum, dataclasses, and Pydantic models, with better error handling and fallback to eval where appropriate. [1] [2] [3] [4] [5] [6]

• API and configuration changes:

  • Changed the return type of DefaultModel property in config.py to OpenAICompatibleModel for better type clarity.

New Features and Examples:

• Added a new example notebook and script for meta-prompt tuning (MetaPromptTuning.py), showcasing advanced prompt engineering and function emulation with Pydantic models.
• Introduced a utility (as_probabilistic) for extracting probabilistic outputs from enum-based model responses.

Documentation and Versioning:

• Updated CHANGELOG.md for v3.0.0 and noted that future changelogs will be maintained in GitHub Releases.
• Bumped version to 3.0.2 in pyproject.toml, __init__.py, and updated the example notebook to use the new version. [1] [2] [3]

---

Pydantic and Type Conversion Enhancements:

• Added dynamic Pydantic class string reconstruction and improved Pydantic model detection and handling throughout the type conversion pipeline. [1] [2] [3] [4]
• Improved robustness and error handling in type conversion functions, especially for Any, Enum, dataclasses, and Pydantic models, with better fallback logic. [1] [2] [3] [4] [5] [6]

New Features and Utilities:

• Added a meta-prompt tuning example with advanced function emulation and Pydantic integration in MetaPromptTuning.py.
• Introduced as_probabilistic utility for extracting probabilistic outputs from LLM enum responses.

Documentation and Version Updates:

• Updated changelog for v3.0.0 and migrated changelog maintenance to GitHub Releases.
• Bumped package version to 3.0.2 in configuration files and updated example notebook version references. [1] [2] [3]

Configuration and API Changes:

• Changed the return type for DefaultModel property to OpenAICompatibleModel for improved clarity.