Add reproduce() function

[mgehre-amd]

• Extract a common model_to_fxgraph() function
• Add a copy of https://github.com/pytorch/pytorch/blob/main/torch/_functorch/fx_minifier.py with fixes to how tuples are handled (will open an upstream PR)
• Add python/torch_mlir/repro.py to reduce torch models while keeping the same error out of torch_mlir.compile(make_fx(model))

Example output:

Strategy: Eliminate dead code (G: 2) (3 nodes, 1 inputs)
FAIL: Eliminate dead code

Strategy: Remove unused inputs (G: 2) (3 nodes, 1 inputs)
FAIL: Remove unused inputs

Strategy: Truncate suffix (G: 2) (3 nodes, 1 inputs)
FAIL: Truncate suffix

Strategy: Delta Debugging (G: 2) (3 nodes, 1 inputs)
FAIL: Delta Debugging
Trying granularity 1

Strategy: Truncate suffix (G: 1) (3 nodes, 1 inputs)
FAIL: Truncate suffix

Strategy: Delta Debugging (G: 1) (3 nodes, 1 inputs)
FAIL: Delta Debugging

Strategy: Remove outputs (G: 1) (3 nodes, 1 inputs)
FAIL: Remove outputs
Made 62 queries
---- SNIP ----
import torch
import torch_mlir

class Model(torch.nn.Module):

    
    
    def forward(self, _to_copy_1):
        cumsum = torch.ops.aten.cumsum(_to_copy_1, 1);  _to_copy_1 = None
        return cumsum
        

model = Model()
model.to(torch.bfloat16)
inps = (torch.ones(torch.Size([1, 12]), dtype=torch.int32), )
out = model(*inps)
torch_mlir.compile(model, inps, output_type='tosa')

---- SNIP ----

[ljfitz]

Super useful - thanks for chasing this down and getting it to work!