Fix num_classes update in reset_classifier and RDNet forward head call

[brianhou0208]

Hi @rwightman ,

1. self.reset_classifier Bug:
   The self.reset_classifier method now properly updates self.num_classes when the num_classes parameter is used. (Only a few models lacked this update, while most already had it correctly implemented)

2. RDNet forward Method Bug:
   Fixed the forward method of RDNet, where it incorrectly called self.head(x). It now correctly calls self.forward_head(x) to ensure proper functionality during the forward pass.

[rwightman]

@brianhou0208 thanks, should probably add a assert model.num_classes == 0 after the model.reset_classifier(0) in test_model_default_cfgs and test_model_default_cfgs_non_std to prevent that in the future...

[brianhou0208]

@rwightman finished it, added assertions to check num_classes after reset_classifier(0) in both tests