[BE] Build PyTorch with `-Wnewline-eof` (#99687)
This would avoid further regressions like the ones reported in https://github.com/pytorch/pytorch/pull/96668#issuecomment-1468029259
Surround some ONNX/flatbuffer includes with `C10_DIAGNOSTIC_PUSH_AND_IGNORED_IF_DEFINED("-Wnewline-eof")` cone of shame
Fixes https://github.com/pytorch/pytorch/issues/96747
Pull Request resolved: https://github.com/pytorch/pytorch/pull/99687
Approved by: https://github.com/kit1980
