+ Store sites in the fundamental domain only
+ Compute and store the symmetry element with each term

Here we continue to build the hopping terms from the graph, however
when extending to handle >1D translational symmetry we will need
to query the Builder directly. We could in principle already do that
at this stage, and it would mean that less has to change later.

Even though we currently import 'Symmetry' and 'NoSymmetry' in
'builder' (to maintain backwards compatibility), we should use
the canonical location within Kwant itself.

N-dimensional systems will need symmetry elements, so the
symmetry definition needs to be in the system module.

Previously we required an array of group elements to act on a
SiteArray (1 element per site), however often we want to act with
a single group element on *all* the sites in a SiteArray.
For TranslationalSymmetry we implement this using array broadcasting.

hamiltonian_submatrix is not meaningful for
infinite systems, and (unlike un-vectorized systems)
is not even used for constructing cell_hamiltonian and
inter_cell_hopping in the vectorized case.

The docstring refers to site orderings, and explicit references to
the graph; these are things that vectorized systems keep for
backwards compatibility, but we wish to move away from. For this
reason we move the docstring to the (unvectorized) infinite system
class and add a more specific docstring to the vectorize infinite
system class.

We now construct the site arrays to pass to 'val' in a single
place (this is more readable).

We had forgotten to pass the 'vectorized' flag when constructing
a Builder with a symmetry

This is a pure refactor.

add plotly plotter

See merge request kwant/kwant!294

Raises a RuntimeError when incompatible inputs are provided to the plot function using the plotly backend.

Called once during setting of plotly backend.

make _normalize_matrix_blocks check number of orbitals

Closes #346

See merge request kwant/kwant!349

This test was previously incorrect, but because we didn't actually
check the number of orbitals on constant values on finalization
this was not caught.

We allow the function that produced the data to be passed to
'_normalize_matrix_blocks', and any error messages are augmented
with this information. This may prove useful for debugging.

Previously these 2 steps were done in one pass. Now we separate them
out in order to facilitate augnmenting the verification to include
checking the number of orbitals (in future commits).

These will be needed later when verifying that constant values have
the correct shape.

It is only used in the tests anyway. This way we export less
API from the builder module.