I think I'm right that we've had our own custom implementation of
Powell for a very long time - it looks to me (from git blame) that
Travis O put a version of the current code in with commit b94c30dcb
(April 2002).
For what it's worth, it's not so much "our own custom implementation" so much as following a particular published variant that was explained (but not originally published in) Numerical Recipes. Interestingly, that chapter claims that the modification is to assist Powell's method in handling long skinny valleys.