- Support hyperspherical normalization
- Support loading FastTD3 + SimbaV2 for both training and inference
- Support (experimental) reward normalization that uses SimbaV2's formulation -- not working that well though
- Updated README for FastTD3 + SimbaV2
