Fix statistic bug in stretch_move

Stretch move returned wrong statistic when rejecting the step. This caused a wired bias
in the resulting distribution.

Kudos to Philipp T. finding this!