We adopt a semi-naive approach. Probably not the best performing and
most robust solution, but at least we have something functional.
We rely on a external scheduler, such as GNU parallel, to concurently
copy the files and use as much cores as possible. It makes the program
composable with the GNU userspace toolchain.
It however comes at a cost: we have to spin-up/tear-down a process for
each file. Scheduling the copy from inside the Rust program by
creating multiple threads (using rayon?) would be likely much more
performant. We'll probably come to it in the future.
The error handling also left to be desired. We're panicking with a
error message every time we encounter something unexpected.
The program is also untested. Might eat your kitten for now.
