native_functions.yaml: reset_grad_accumulator (#36431)
Summary:
Add optional 'reset_grad_accumulator' in native_functions.yaml
Use it for all overloads of `set_`
Fixes https://github.com/pytorch/pytorch/issues/33941
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36431
Differential Revision: D21156062
Pulled By: ezyang
fbshipit-source-id: 5a63fd091a618a33cb05fd96bbb1e87162abc9a4