Feature request #2129: Use map-reduce approach to make Command-T faster

Command-T is mostly CPU-bound, and well suited to a map-reduce approach.

The matching algorithm matches and scores the set of paths in the search space, then sorts the result by score. It would be relatively straightforward to divvy up the search space across the CPU cores, then merge the lists and do the sort.

The only problem is that Ruby multithreading isn't real. This ticket, then, is to capture the idea of a branch in which we use POSIX threads to do this "outside" of Ruby. That means it won't work on Windows (which is why it must be a separate branch), and also that I might have to replace some of the internals with hand-coded data structures (don't want any use of Ruby objects like strings and arrays to potentially invoke garbage collection, which is basically certainly not thread-safe).

Not sure if this is viable, but wanted to note it down as a possibility before I forget it.