Catch invalid scalar type when parsing tensors (#1518)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/1518

Fail non-fatally when encountering an unknown/unhandled `ScalarType` in a `.pte` file. As part of this:
- Move the "types not supported yet" logic out of `scalar_type_util` and into `tensor_parser`, since that decision is an aspect of the runtime and not a fundamental aspect of `ScalarType`.
- Remove the now-duplicate `sizeof_scalar_type` function, which is the same as the exsting `elementSize` function. Before this diff, `sizeof_scalar_type` did the "unsupported" checks that have now moved.
- Add an `isValid()` function to let users of `ScalarType` know whether a given enum value is legit. This makes it possible to avoid the fatal error when calling `elementSize` on a bad value.
- Add unit tests for the new `isValid()`.

Reviewed By: larryliu0820

Differential Revision: D52451738

fbshipit-source-id: 88e47d7a3c688e3e2a68cc86114935c7ff6a73b5