Move etensor types to their new namespace (#5569)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/5569

Move these types out of the old `torch::executor` namespace. Put them in their own namespace so that no-one will see them by default, unlike their old location in `torch::executor`.

Most users should not use these directly, instead using the `executorch::aten` aliases. These ETensor types live under `executorch::runtime...` to make it a little more clear that they're semi-internal.

Also:
- Hide bfloat16's `bits_from_f32`, which was only used by tests
- Hide the `min` implementation used by string_view. Although it was in an anonymous namespace, it was in a header, so it was visible to all code in `torch::executor`.
- Remove an unnecessary forward declaration from BlasKernel.h which conflicted with the new namespace

Reviewed By: swolchok

Differential Revision: D63294217

fbshipit-source-id: 5241fd5a0b487631b6460222b6c3e27e82aea5f0