DF with Kállay's trick

[susilehtola]

There's a big difference in speed between MemDFJK and DiskDFJK. For the former case, it would probably not be a bad idea to compress the DF integrals with Kállay's trick from J. Chem. Phys. 141, 244113 (2014), which can reduce the number of auxiliary functions. I assume the savings would be especially significant for small basis set calculations like 6-31G*.