Properly register quantized out ops into AOT export flow (#3449)

Summary:
Currently are registering all ops specified in quantized.yaml into AOT, that causes double registration problem because some of them are already registered in https://github.com/pytorch/executorch/blob/main/exir/passes/_quant_patterns_and_replacements.py#L185-L188

This PR changes the way we do selective build for these quantized ops, by listing out individual ops that we want to register.

We also updated the current CI job to be able to error out when this registration doesn't work.

Pull Request resolved: https://github.com/pytorch/executorch/pull/3449

Reviewed By: mergennachin

Differential Revision: D56844556

Pulled By: larryliu0820

fbshipit-source-id: 7b77aa7e019543cbdcce02b6f6662acf31dbb3f4