Add t-test with unequal variances, bug fix for edge case on all tests

[MrCreosote]

Added method for t-test with unequal variances. ttest_ind assumes
population variances are equal.

Refactor common code in all 4 t-test methods to _ttest_ind_setup and
_ttest_finish.

Fix bug in all 4 methods. Previous implementation would, if the
difference in means and the denominator of the t-statistic were zero,
set the t-statistic equal to 1. If the difference in means is zero, the
t-statistic should always be zero. Code and respective tests changed to
reflect this.

[MrCreosote]

Gmane discussion here.

[pv]

Note that you can git push -f to the same branch, if you need to rebase/amend a commit --- no need to open a new pull request.