A cleaner and more scalable implementation of symbolic tracing (#11763)

Cleaner and more scalable implementation of symbolic tracing with torch.fx, and provides support for new architectures:
- ALBERT
- DistilBERT
- MobileBERT
- MegatronBERT
- GPT2
- GPT Neo

Co-authored-by: Michael Benayoun <michael@huggingface.co>

tests/test_modeling_gpt_neo.py

@@ -276,6 +276,7 @@ class GPTNeoModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase
 
     all_model_classes = (GPTNeoModel, GPTNeoForCausalLM) if is_torch_available() else ()
     all_generative_model_classes = (GPTNeoForCausalLM,) if is_torch_available() else ()
+    fx_ready_model_classes = all_model_classes
     test_missing_keys = False
     test_pruning = False
     test_model_parallel = False