[Static Runtime] Avoid evaluating expressions of `Node*` for interpreter fallback op (#69489)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/69489
This change avoids pulling out `Node*` out of `ProcessedNode*` to evaluate expressions related to `Node*` at op execution time.
Perf gain is expected to be there but not measurable and the purpose of this change is to make SR's code more self-contained (calling more code from SR not JIT) during execution time.
Test Plan: Existing tests
Reviewed By: mikeiovine
Differential Revision: D32893265
fbshipit-source-id: f0f397666b3556f985d45112af8fe0b08de22139
