add output_process_fn_grad before sum().backward() (#59971)
Summary:
This should fix `to_sparse` test issue.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/59971
Test Plan:
CI
Also: directly examine the RuntimeError thrown from test_unsupported_backward
- Before:
```
NotImplementedError: Could not run 'aten::sum' with arguments from the 'SparseCPU' backend.
```
- After:
```
to_dense() not supported for float16 on CPU
```
Reviewed By: soulitzer
Differential Revision: D29112558
Pulled By: walterddr
fbshipit-source-id: c2acd22cd18d5b34d25209b8415feb3ba28fa104
Author
Rong Rong (AI Infra)
