docs: connect DPO data, model, and training code

data: finalize canonical preference release

Add files using upload-large-folder tool

initial commit