Fix torch::nn layers to always subclass from `torch::nn::Cloneable` (#27770)
Summary:
The impl class of `torch::nn` layers must always subclass from `torch::nn::Cloneable`, otherwise `module->clone()` doesn't work on them. This PR fixes layers that don't conform to this rule.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/27770
Differential Revision: D17893051
Pulled By: yf225
fbshipit-source-id: 37cdf8c09e22f0f164cbd0e8700965a1778ec4c1
