Enhance DataFrame.collect to utilize Rayon for parallel RecordBatch conversion

- Added Rayon as a dependency in Cargo.toml.
- Implemented a benchmark script (collect_gil_bench.py) to measure performance of serial vs parallel conversions.
- Updated documentation (collect-gil.md) to explain the impact of GIL on performance and how to run the benchmark.
- Modified the collect method in PyDataFrame to release the GIL and convert RecordBatches to PyArrow in parallel, improving CPU utilization.

Author: kosiew