BASALT: A Benchmark For Learning From Human Suggestions