Various performance enhancements

[MichaelSchreier]

• Replaced calls to numpy.<ufunc>(numpy.ndarray) with numpy.ndarray.<ufunc>(). The latter circumvents a check whether the input is a numpy.ndarray and is particularly effective for relatively small inputs. But there are still notable time savings even for very large arrays.
• Replaced itertools.permutations with itertools.combinations as the distance d between any two points A and B must only be computed once rather than as d(A,B) and d(B,A). There are a few extra steps necessary to then fill out the distance and index arrays so the scaling is quite a bit shy of the 50% one might expect at first glance.

With the real datasets that I'm working with the first change alone boosts performance by almost 30% and the second one by another 35% for an overall decrease in runtime by about 50%.

[coveralls]

Pull Request Test Coverage Report for Build 62

• 13 of 13 (100.0%) changed or added relevant lines in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.08%) to 95.588%

---

Totals
Change from base Build 61:	0.08%
Covered Lines:	260
Relevant Lines:	272

---

💛 - Coveralls

[coveralls]

Pull Request Test Coverage Report for Build 62

• 13 of 13 (100.0%) changed or added relevant lines in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.08%) to 95.588%

---

Totals
Change from base Build 61:	0.08%
Covered Lines:	260
Relevant Lines:	272

---

💛 - Coveralls

[vc1492a]

@MichaelSchreier thanks for the PR! Appreciate the effort, I'll incorporate the changes into a future release. I've updated the PR to merge into dev instead of master. Thanks!