Command execution judges for Agent Judge - build, test, and script validation

Core abstractions, runner, comparison, and results for experiment driver

Core benchmarking framework for evaluating Java-focused AI agents