In my previous NaN-related commit, I added a conditional similar to the
one seen in glMatrix, but this was insufficient for the interpolation
seen in a glTF skinning test because a divide-by-zero was occuring later
in the function. This commit simply adds back the previous protection.