Work in parallel

[Noratrieb]

cargo-minimize is pretty embarassingly parallel. It could copy the soruce to a bunch of tempdirs and run in parallel there, then atomically committing the minimized results into the real dir (this would be little tricky to get right but entirely possible).

Parallelization would be a great speedup as most performance is lost when bisecting through large "mostly no" files. As few things get committed during that work, it would be very easy to parallelize.