docs: point at the packaged eval suites and split-hygiene module

docs: correct cross-rendering figure after tier-classifier fix

docs: dataset card with benchmark caveat and provenance

Upload dataset

initial commit