[xnnpack][on-device] compiler --> executor object (#88779)
#### XNN Compiler Object
This is purely to abstract away the subgraph rebuild from the flatbuffer object. CompileModel return an executor object which we can use to setup inputs and run forward with.
#### Executorch Considerations
We Include ATen/utils for torch_check, this will be changed when moving to executorch
Differential Revision: [D40733163](https://our.internmc.facebook.com/intern/diff/D40733163/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/88779
Approved by: https://github.com/digantdesai
