Optimize mean_change (fixes issue #542) and correct documentation

[blu3r4y]

This uses a simpler formula for calculating mean_change that was already proposed in #542. The proof for this simplification follows. Moreover, I fixed the documentation, because we actually divide by n - 1 because a series with n elements only has n - 1 differences. I also added a few test cases to check some corner cases.

[coveralls]

Coverage remained the same at ?% when pulling 955479f on blu3r4y:optimize-mean-change into c3d35c4 on blue-yonder:master.

[dbarbier]

Likewise mean_second_derivative_central could return (x[-1] - x[-2] - x[1] + x[0]) / (2 * (len(x) - 2))

[nils-braun]

Makes sense to do it like this! Thanks.
@dbarbier do you want to include the fix you mentioned as another PR?

[dbarbier]

Okay, will do, thanks for merging these PRs.