Expand benchmark evidence with functional and inverse-folding metrics

Publish dual-learning showcase with live inference and sourced annotation cases