[fx2trt] Use weights shape as normalize shape in layer norm (#69401)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/69401
As the title. In PyTorch, these two shapes are the same. Normalize shape might be retrieved from tensor.size and in explicit batch dim, it won't work right now.
Test Plan:
```
✓ ListingSuccess: caffe2/test/fx2trt/converters:test_layernorm - main (7.018)
✓ Pass: caffe2/test/fx2trt/converters:test_layernorm - test_layer_norm_with_dynamic_shape_0_1d_normalized_shape (caffe2.test.fx2trt.converters.acc_op.test_layernorm.TestLayerNormConverter) (22.945)
✓ Pass: caffe2/test/fx2trt/converters:test_layernorm - test_layer_norm_0_1d_normalized_shape (caffe2.test.fx2trt.converters.acc_op.test_layernorm.TestLayerNormConverter) (23.203)
✓ Pass: caffe2/test/fx2trt/converters:test_layernorm - test_layer_norm_with_dynamic_shape_1_2d_normalized_shape (caffe2.test.fx2trt.converters.acc_op.test_layernorm.TestLayerNormConverter) (42.549)
✓ Pass: caffe2/test/fx2trt/converters:test_layernorm - test_layer_norm_1_2d_normalized_shape (caffe2.test.fx2trt.converters.acc_op.test_layernorm.TestLayerNormConverter) (43.544)
✓ Pass: caffe2/test/fx2trt/converters:test_layernorm - test_layer_norm_with_dynamic_shape_2_4d_input_shape (caffe2.test.fx2trt.converters.acc_op.test_layernorm.TestLayerNormConverter) (45.958)
✓ Pass: caffe2/test/fx2trt/converters:test_layernorm - test_layer_norm_2_4d_input_shape (caffe2.test.fx2trt.converters.acc_op.test_layernorm.TestLayerNormConverter) (47.027)
Summary
Pass: 6
ListingSuccess: 1
```
Reviewed By: yinghai
Differential Revision: D32853359
fbshipit-source-id: 8a122fe3348a1d9ad07b48647ec6166d171d113a