Port lu_solve to ATen (#22379)
Summary:
Changelog:
- Port TH implementation to ATen/native/BatchLinearAlgebra.cpp
- Port THC implementation to ATen/native/cuda/BatchLinearAlgebra.cu
- Remove TH/THC implementations
- Update doc strings
Pull Request resolved: https://github.com/pytorch/pytorch/pull/22379
Test Plan: - Added new tests in test_torch.py (port to test_cuda.py exists)
Differential Revision: D16089645
Pulled By: zou3519
fbshipit-source-id: dc8561aadacacb23e80c375b4fec687df2b6bbc8