Kernel Bench + torch.compile

[rengolin]

Add the torch.compile to kernel_bench as an option to test both import types when running a pipeline with it.

Unify this:

• https://github.com/llvm/lighthouse/blob/main/tools/kernel_bench

With this:

• https://github.com/llvm/lighthouse/blob/main/examples/ingress/torch/compile_model.py