perf: prepare to handle multiple pairs in perfbdiff Before, we only supported benchmarking a single pair of texts with bdiff. We want to enable feeding larger corpora into this benchmark. So rewrite the code to support that.