Fix Spearman's rank correlation to address #2. Remove tiedrank method with dimension argument. Enable tests.

[andreasnoack]

I also think we should remove the cov_spearman methods. Are you aware of any applications?

[johnmyleswhite]

Looks good to me. I'm not aware of any use cases for the rank order covariance.

[andreasnoack]

I have now removed Spearman rank covariance functions and added Kendall's rank correlation functions. Please notice that the latter is a rather fast implementation. For two vectors of length 50 000

MATLAB:

>> tic;corr(a,b,'type','Kendall');toc
Elapsed time is 13.560623 seconds.

R:

> system.time(cor(a,b,method='kendall'))
  bruger   system forløbet 
  60.241    0.079   60.642 

Julia:

julia> @time cor_kendall(a,b);
elapsed time: 0.421725324 seconds (40951872 bytes allocated)

[johnmyleswhite]

Sweet.

[johnmyleswhite]

I say we merge this.