Improve statistics computation for different time step strides

[LTMeyer]

Context

Related to #44 (comment) of #44 and #41.

Delta statistics have been computed on one time step difference between snapshots. They are thus incompatible for different strides. We currently raise an error if the stride is incompatible with the normalization. But we may want to offer better normalization approach.