Performance improvement for compute_enpt2 and add_hci

[peter-reinholdt]

Some work can be avoided when searching for determinants in the compute_enpt2 and add_hci routines.
Rewrite the criterion

$$ \left|H_{ki} c_{i} \right| > \varepsilon $$

as

$$ \left|H_{ki}\right| > \frac{\varepsilon}{\left|c_{i}\right|} $$

The value of the Hamiltonian matrix element is independent of what happens with, e.g., excite_det - so sometimes work can be skipped.

The performance improvement is system-dependent, but here is an example for water/cc-pVDZ.
Timings are in seconds, single thread. The benefit is most substantial when the majority of the determinants are rejected; when most determinants are accepted, there is almost no benefit.

eps1	eps2	Routine	Before	After	Speedup
1.00E-03		add_hci	0.093	0.042	2.23
5.00E-04		add_hci	0.253	0.168	1.51
2.00E-04		add_hci	0.888	0.416	2.14
1.00E-03	0.0001	compute_enpt2	0.237	0.163	1.45
1.00E-03	1.00E-05	compute_enpt2	0.773	0.613	1.26
1.00E-03	1.00E-06	compute_enpt2	2.167	1.913	1.13
5.00E-04	5.00E-05	compute_enpt2	0.720	0.495	1.46
5.00E-04	5.00E-06	compute_enpt2	2.070	1.731	1.20
5.00E-04	5.00E-07	compute_enpt2	8.462	8.005	1.06
2.00E-04	2.00E-05	compute_enpt2	3.097	2.014	1.54
2.00E-04	2.00E-06	compute_enpt2	9.319	7.954	1.17
2.00E-04	2.00E-07	compute_enpt2	46.647	45.395	1.03

[PaulWAyers]

Thanks @peter-reinholdt . I assigned Michelle to look at this. You should be aware that we are doing a major rewrite of PyCI right now, so we are in "maintenance mode" on the current public version of PyCI right now. That said, to my naive eyes, this should be easy to transfer over to the new back end.