Categorize model failures in generating shlo graph

[AleksKnezevic]

We have some ops in the model test that either can't generate torch-mlir (status=2 or 3) or convert it to stable HLO(status=4), see tracker

Can you please investigate what's causing the failures?

The first step will be to add try expect blocks around the section that fails and log the errors into the generated json. Once this is ready, let's discuss next steps.