Improve efficiency of lehvenshtein_distance method

[eendebakpt]

The buffer initialization (which happens many times) is improved by replacing the multiplication with a sum.

[JelleZijlstra]

Have you benchmarked this? MOVE_COST is 2, so the multiplication should just be a bitshift. Chances are that's faster than an additional memory read.

[eendebakpt]

@JelleZijlstra Fair point. As long as MOVE_COST is a power of two the compiler can indeed optimize and there is no gain. I benchmarked on my system with MOVE_COST=2 (just the buffer filling in a large for loop):

current main: 1.48 seconds
original PR: 1.55 seconds
updated PR: 0.89 seconds