Improving npairs(s, mu) performance and testing.

[manodeep]

Completely un-tested, un-benchmarked. I am assuming there are existing tests for the npairs(s, mu) pair counters.

Related to #763

[duncandc]

Performance is about 10x's worse for me when applying this code. Are there some compiling flags that we need implement?

[aphearin]

Cython's automated compiler flags are sufficiently good that a 10x improvement never occurs by simply adding flags. If you just run cython -a on the module, it showed that there were a couple of forgotten declarations that I just added just now.

[duncandc]

Excellent. That did the trick. It looks to be of order 50% performance increase over the code in the repo right now. Im going to push my modifications to the pair counters and tests to this PR.

[duncandc]

Okay, I plan on merging this with master after it passes the travis build. I think the brute force pairs.s_mu_npairs comparison test should alleviate any concern that may have been lingering about the reliability of the optimized npairs_s_mu() function. I have also edited the doc strings of the s_mu pair counters.

[duncandc]

@aphearin do you expect some travis builds to fail currently?

[duncandc]

okay, note that doc strings with math need to be preceded by an 'r' in order to pass python 3 builds.