FIX: rank-size test by inc. sample size

[bktaha]

As mentioned in PR #551 the test for rank_size method checks for linearity in the output log-log plot. I did this by performing linear regression of log(size_data) on log(rank_data) and comparing the resulting r_squared value to 1.

Turns out an error tolerance of 1e-4 for a sample size of 1000 was a little optimistic. I have increased the sample size to 10000, reduced the tolerance to 1e-3 and the comparison works as expected now.

Another mistake was setting the seed for exponential draw but not the Pareto draw, I've fixed that as well. I'm not sure if setting a seed here is a good practice in the first place, if not I can remove it and increase sample size / reduce tolerance further.

Please let me know if this is an acceptable fix. This should close #553.

[mmcky]

thanks @bktaha for this well documented PR and for looking into the test failure. It is greatly appreciated.
@jstac would you mind to cast a quick eye over this if you have time?

[jstac]

Nice work @bktaha. Many thanks. This looks good to me @mmcky.

[mmcky]

thanks @jstac