The merge operation takes tensors, finds the union hypercube (ie
the hypercube that contains all the input tensor hypercubes) and
spreads the tensors as necessary to make them conformant. Finally,
the resultant tensor has an additional string dimensioned axis, each
element of which is one of the input tensors. This dimension should be
named using the operation edit dialog. It is an error for a hypercube
to contain more than one axis with the same name.