[proposal] Use self.flatten instead of torch.flatten and when becomes possible derive ResNet from nn.Sequential (scripting+quantization is blocker), would simplify model surgery in the most frequent cases

Currently In https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py#L243:

        x = self.avgpool(x)
        x = torch.flatten(x, 1)
        x = self.fc(x)

If it instead used x = self.flatten(x), then it would simplify model surgery: del model.avgpool, model.flatten, model.fc. Also in this case the class can just derive from Sequential and use OrderedDict to pass submodules (like in https://discuss.pytorch.org/t/ux-mix-of-nn-sequential-and-nn-moduledict/104724/2?u=vadimkantorov), this would preserve checkpoint compat as well. The method forward could then be removed

Start in torchvision/models/resnet.py around line 243 and review the linked PyTorch discussion about mixing nn.Sequential and nn.ModuleDict. Investigate how scripting and quantization constrain replacing torch.flatten and deriving ResNet from Sequential, including checkpoint compatibility. Done means the proposed model surgery is supported without breaking those constraints.