OK, apparently I get competitive when someone else finds an astcenc optimization that I missed 😂

Added enough coffee and found some nice opts for one of the gnarly hot functions I've been battering my head against for years. Even managed to go bouldering while tests were running.

One nice caching optimization to avoid recomputing infill weights every texel (precompute set, and then patch the set if things change).

A couple of analytical error term alterantives that reduce in-loop maths.