[export][training ir migration] quantized_decomposed.quantize_per_tensor decomposition (#134525)

Summary:
In graph of  TestXNNPACKQuantizer.test_dynamic_linear_with_con test, some quantized_decomposed.quantize_per_tensor.default ops are becoming quantized_decomposed.dequantize_per_tensor.tensor ops when using the new training ir.

This is because we lift params/buffers before calling make_fx. So previously, for the graph that’s passed to make_fx,`graph.L__self___linear1.weight` is a tensor
now in training ir, graph.L__self___linear1.weight is a FakeTensor. This caused the node overload to be different.

Test Plan:
```
buck2 run 'fbcode//mode/dev-nosan' fbcode//caffe2/test/quantization:test_quantization -- -r test_dynamic_linear_with_conv
```

Differential Revision: D61364547

Pull Request resolved: https://github.com/pytorch/pytorch/pull/134525
Approved by: https://github.com/tugsbayasgalan, https://github.com/jerryzh168

torch/testing/_internal/common_quantization.py

@@ -1247,6 +1247,7 @@ class PT2EQuantizationTestCase(QuantizationTestCase):
         export_with_dynamic_shape=False,
         is_qat=False,
         is_debug_mode=False,
+        capture_pre_autograd_graph_node_occurrence=None,
     ):
         # resetting dynamo cache
         torch._dynamo.reset()
@@ -1305,6 +1306,10 @@ class PT2EQuantizationTestCase(QuantizationTestCase):
             for k, v in PT2EQuantizationTestCase._MAP_TO_FX_TRACED_OPS.items():
                 if k in expected_node_occurrence:
                     node_occurrence[ns.call_function(v)] = expected_node_occurrence[k]
+            if capture_pre_autograd_graph_node_occurrence is not None:
+                node_occurrence = {
+                    ns.call_function(k): v for k, v in capture_pre_autograd_graph_node_occurrence.items()
+                }
             self.checkGraphModuleNodes(m_fx, expected_node_occurrence=node_occurrence)
             fx_quant_output = m_fx(*example_inputs)
             self.assertEqual(fx_quant_output, pt2_quant_output)