perf: walk with tidy_walk

The CLI's walk and thread pool move to the shared tidy_walk crate,
which reads, formats and writes in one pass and lists each file once,
however it is spelled. One pool of core-count threads beat the macOS
pipeline default here: A+B 39.8 -> 33.8 ms, C 69.8 -> 54 ms.

tidy_walk is a path dependency on ~/repos/oss/tidyrs, which has no
remote yet, so CI can't build this workspace until that changes.