DTR: Speed up generation

[bgpeyton]

Previously issue #30 from lml.

A few hold-overs from the last DTR optimization PR. Would like to try these to speed up DTR generation when the number of density elements exceeds a few hundred (often the case when cut_type == max or AO-basis densities are used).

• Reject density elements d such that g(x,d,s) == 0 to within machine precision
• Option to parallelize sums?