LinearEinsum: fix adjoint_times for pure transpose operations

Lukas Scheel-Platzrequested to merge
fixup_einsum_transpose into NIFTy_6

LinearEinsum, when called with an empty mf and subscripts 'ij->ji' did calculate incorrect adjoint subscipts ',yz->zy', rendering its .adjoint_times method inoperable.

Fixed this and added a test on it.

@gedenhof, could you please check if the fix is really correct?

Merge request reports