Prevent audit artifacts from breaking dataset viewer config

Fix dataset viewer config to use one Parquet eval split

Make eval viewer schema order explicit

Make eval viewer schema order explicit

Put grapheme and phonetic first in eval rows

Add frequency-preserving surname phonetic G2P evaluation dataset

initial commit