Benchmarking and Building Long-Context Retrieval Models with LoCo and M2-BERT