perf: 6 io threads on macos

mdtidy measured its corpora, which have fewer, larger files and more
work per byte, and 4 I/O threads were up to 18% behind the best
shape there. Sweeping the real crate with tidy's and mdtidy's
formatters, 6 I/O threads with workers on every core is never more
than 9% behind the best shape, where every other shape loses 18 to
42% somewhere.