adcg-compression

ADCG works now, but the local improvement step is slow. This commit includes some debugging code that I anticipate needing for awhile.

Interior point method for solving the convex step of ADCG. Attempted to improve performance of gradient computations but mostly failed.

Switch to numerical gradients for now. This commit contains a bunch of debugging code.

AdcgEncoder kinda sorta works, ish. (NLO for gradient computation does something, but it probably needs to be tuned.)

Initial work on ADCG encoding, with substantial refactoring. ADCG encoder is still under construction.