Use int8 quantizer in the OSS flow (#6166)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/6166

As titled. This change add the ability to supply a qconfig to the `CadenceQuantizer`, and uses `int8` instead of `uint8` in `export_model` as per Cadence request. Going forward, `int8` should be the primary 8-bit type.

Reviewed By: dulinriley

Differential Revision: D64209639

fbshipit-source-id: c8bb385aa75cdeb0cfb92217f4cdc5335a10a3b9