Switch benchmarking to use export non-strict mode (#130977)
Summary:
Switch the export part used by AOTInductor benchmarking from strict to non-strict, and switch it from producing torch IR to aten IR.
X-link: https://github.com/pytorch/pytorch/pull/130977
Approved by: https://github.com/angelayi
ghstack dependencies: #134639
Reviewed By: ZainRizvi
Differential Revision: D62009176
Pulled By: desertfire
fbshipit-source-id: 72cb7b2f4ef8297537a932f30010ebcb46e12935
