Add rank-based statistics to Series.correlation()

[CommonClimate]

Currently, Series.correlation() only supports Pearson's product-moment correlation as metric of similarity. This is the only possible statistic for the 'ttest' option. However, the two Monte Carlo tests (isospectral and isopersistent) can use any statistic, particularly Spearman's rho and Kendall's tau, which are rank-based and therefore more resistant to outliers.

The idea is to use the same AR(1) or phase-scrambled surrogates (see #228) to general the null distribution. For a student project, it may be useful to validate those against the asymptotic distribution of those statistics.

[CommonClimate]

Update: it would make sense to enable any of the correlation methods supported by scipy, using the same nomenclature. Example:
corr = ts1.correlation(ts2, statistic = "kendalltau") or corr = ts1.correlation(ts2, statistic = "spearmanr").
Default statistic should be 'pearsonr'. The function will need a check that the supplied string matches the list of allowable methods under scipy.stats.

[khider]

Depends on solving issue #228