The BiGGen Bench: A Principled Benchmark for Fine-grained Evaluation of Language Models with Language Models