Make scan parallel

[faern]

unicop is currently single threaded. This makes scanning large code bases a lot slower than it has to be. Make it multi-threaded one way or another.

One suggestion is to keep the directory traversal single threaded in one place. And for each found file the path is sent into some pool of worker threads. These worker threads pick up the message, reads the file and perform the unicode validation on it. The worker thread then returns the result on a channel, instead of directly printing the errors. A special reporter thread reads from the worker threads and prints it.

This should be made in such a way that the output order of errors is deterministic and always the same for a given input code base. The errors should be reported in the same order as the files are traversed on the disk.