Update aot_arm_compiler to use export_for_training (#5581)

Summary:
Remove capture_pre_autograd_graph and use updated export_for_training within the aot_arm_compiler.py script

Pull Request resolved: #5581

Reviewed By: kirklandsign

Differential Revision: D63367930

Pulled By: digantdesai

fbshipit-source-id: 3b161432d533d8debcf7fdfdb244b69fa48adc8d

Author: SaoirseARM

examples/arm/aot_arm_compiler.py

@@ -288,7 +288,7 @@ def get_args():
     model = model.eval()
 
     # pre-autograd export. eventually this will become torch.export
-    model = torch._export.capture_pre_autograd_graph(model, example_inputs)
+    model = torch.export.export_for_training(model, example_inputs).module()
 
     # Quantize if required
     if args.quantize: