corrected some conceptual mistakes in StandardHamiltionian that had no effects because the prior metric is 1

found a bug in linearization where __getitem__ would adjust the value correctly but the jacobian incorrectly

This behaviour is not intended according to the docstring.

It is rendered in the docs before domain.shape and as such hard
to understand, because it explains two concepts (shape, MPI data
splitting) at once.

Therefore I added the hint that it is mainly relevant for MPI
and that the docstring of domain.shape explains the shape concept
in general.