Defining type 2 diabetes polygenic risk scores through colocalization and network-based clustering of metabolic trait genetic associations