Optimize error checking in mvlgamma (#32665)
Summary:
- Clean up error checking code
- Avoid unecessary floating-point computation
- Use float instead of double when possible to avoid massive cast in the tensor
- Use bool instead of uint8_t for clear Boolean purpose
- Improve error message
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32665
Differential Revision: D19601920
Pulled By: VitalyFedyunin
fbshipit-source-id: 0c6c6b5ff227b1437a6c1bae79b2c4135a13cd37
