[PyTorch] Use uint32_t for ProcessedNode::num_outputs (#121335)
We already use uint32_t for indexing, and the notion of a single graph node with more than four billion outputs stretches credulity.
Differential Revision: [D54598821](https://our.internmc.facebook.com/intern/diff/D54598821/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/121335
Approved by: https://github.com/Skylion007
