New Features

Support for AWS Bedrock backend for llm_generate and embed.
Support passing user defined functions that return scalars to groupby.agg and groupby.apply
Support renaming DataFrame column using df.columns = [...] syntax
Add API map_partition_with_state to DataFrame that allows you to do a one-time initialization of state on each worker which can then be used to map batches of rows from a DataFrame to produce a new DataFrame.
Added JIT fallback to Bodo DataFrames such that operations not supported natively in DataFrames can use the equivalent operation from Bodo engine.

Performance Improvements

Improve Series.quantile/describe performance.
Improve the performance of fetching row counts for Parquet datasets
Improve package import time and worker spinup time substantially

Bug Fixes

Fix a crash with llm_generate and embed in Jupyter Notebooks/when an asyncio executor is already running.
Fix OpenAI environment variables not being sent to workers.
Fix bug in loss computation when fitting LogisticRegression in parallel.
Fix crash when running map/apply on large numbers of workers