Restructure as DMPKBench with ADMET-MPO and pk-interpretation subsets (#3)

Convert questions to open-answer format (#2)

Ask agents to explain their reasoning

Add DMPKBench result interpretation subset