Remove inline Tensor decl from tensor_util.h (#4577)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/4577

Header files must not use `using` declarations to create aliases, especially just to reduce typing. https://google.github.io/styleguide/cppguide.html#Aliases

In this case it added extra confusion because it made `exec_aten::Tensor` available as `torch::executor::Tensor`; but, `torch::executor::Tensor` is the "lean" implementation of tensor, and `exec_aten::Tensor` is an alias for `torch::executor::Tensor` in lean mode, but not in aten mode.

Differential Revision: D60869599

fbshipit-source-id: aaa8df51e3eac4ee33d8b6c863d08e7cf6a3eeb8