[StaticRuntime] Add out-variant support for TensorExprDynamicGroup op (#69479)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/69479
This diff adds support for out-variant optimization for `TensorExprDynamicGroup` op, which will be used for TensorExpr based fusion in Static Runtime.
ghstack-source-id: 146107008
Test Plan:
```
buck run mode/opt //caffe2/caffe2/fb/predictor:pytorch_predictor_test
```
Completed accuracy test on inline_cvr model 294738512 v0. Results:
```
get 1012 prediction values
get 1012 prediction values
pyper_inference_e2e_local_replayer_test.out.132ea03c2 pyper_inference_e2e_local_replayer_test.out.1858bbeb0
max_error: 0 % total: 0
```
Reviewed By: d1jang, mikeiovine
Differential Revision: D32768463
fbshipit-source-id: a3e6c1ea9ff5f3b57eb89095aa79a6d426fbb52a
