Rewrite find_bytestr for faster code

Rewrite find_bytestr for faster code

One of the things that have always complicated the main search loop
was the effort to keep forward and reverse search in one loop. It
turns out that this prevents a number of useful optimizations, so
now after the algorithm is well tested, let's separate them.

There's not as much code duplication as one would expect, because I
already had to differentiate between forward and backward in many
places...