Removes unnecessary no_pretrained_model from test_quantize_fx.py (#67836)

Summary:
TorchVision accidentally included model builders for quantized models without weights; this was an old bug. These builders were largely unusable and caused issues to the users. Commonly they were filtered out to avoid causing issues.

We've recently fixed that (pytorch/vision#4854) by either removing those unnecessary builders or by providing quantized weights. This PR removes the no-longer necessary filtering of the methods.

**It should be merged after TorchVision is synced on FBCode.**

Pull Request resolved: #67836

Reviewed By: jerryzh168

Differential Revision: D32230658

Pulled By: datumbox

fbshipit-source-id: 01cd425b1bda3b4591a25840593b3b5dde3a0f12

Author: datumbox

test/quantization/fx/test_quantize_fx.py

@@ -5605,11 +5605,9 @@ def get_available_classification_models(models):
         model_list = get_available_classification_models(models)
         quantized_model_list = get_available_classification_models(quantized_models)
 
-        no_pretrained_model = set(['shufflenet_v2_x0_5', 'shufflenet_v2_x1_5', 'shufflenet_v2_x2_0'])
-        quantized_model_list = set(quantized_model_list) - no_pretrained_model
+        quantized_model_list = set(quantized_model_list)
         # test eager and graph consistency
         model_list = quantized_model_list
-        model_list = set(model_list)
         # mobilenet/inception_v3/googlenet qat is not working due to AdaptiveAveragePool qat
         # we might observe the output of AdaptiveAveragePool in the future
         # and re-enable the test