feat: add batchnorm kernel

[0xrushi]

❯ python main.py batchnorm_benchmark --benchmark=True
Running BatchNorm benchmark: 512x2048, use_triton=True
Output shape: torch.Size([512, 2048]), dtype: torch.float32
Benchmark results
                             kernel_path  non_triton    triton  non_triton-triton
kernel                                                                           
batchnorm_benchmark  batchnorm_benchmark     0.00917  0.000153           0.009017

==================================

❯ python main.py batchnorm_benchmark --benchmark=True --batch_size=1024 --channels=4096
Running BatchNorm benchmark: 1024x4096, use_triton=True
Output shape: torch.Size([1024, 4096]), dtype: torch.float32
Benchmark results
                             kernel_path  non_triton    triton  non_triton-triton
kernel                                                                           
batchnorm_benchmark  batchnorm_benchmark    0.008819  0.000216           0.008603

==================================

[adamomainz]

This really seems more suited for tritonbench
https://github.com/meta-pytorch/tritonbench

If you move it there I would be happy to review