Allow mode-space solvers to work with systems that have self-energy leads attached

Closes #368

See merge request kwant/kwant!366

We cannot, of course, calculate scattering wavefunctions that
*originate* in a selfenergy lead, as that would require knowing
the modes. For this reason the test was changed from checking
for 'NotImplementedError' to 'ValueError'

This transformation is a pure refactor. It was not done in the previous
commit so that the relevant change would be clear.

the weird 'if: x; continue' idiom is just to keep the changeset
small and relevant. The next commit will fix this by indenting
the existing code into an 'else' clause.

It is meaningful to calculate scattering matrix elements for
leads that are specified as modes, but not for leads specified
as selfenergies only.

smatrix should "just work" as long as we are not
requesting scattering amplitudes to/from a selfenergy lead.
Similarly wave_function should "just work" as long as we
are not asking for a wavefunction originating from a
selfenergy lead.

Add a clarification to the self-energy calculation for leads

See merge request kwant/kwant!353

Vectorize wraparound

See merge request kwant/kwant!363

Fix trans symm act

See merge request kwant/kwant!364

fix builds against mumps when using conda/nonstandard shells

Closes #375

See merge request kwant/kwant!362

If the shell has not been set up correctly when using Conda,
the necessary compiler flags will not have been set. We explicitly
set them to avoid this problem.

Invoking "gcc" ourselves does not necessarily invoke the compiler
with the same flags that will be used when the extensions are
actually compiled.

We support Python 3.6, hence >=3.6

remove deprecated matplotlib API from plotter

Closes #366

See merge request kwant/kwant!361

We have to modify all call-sites to use tuple unpacking

It turns out that Matplotlib colormaps are just callables that
map the interval [0, 1] to a line in colorspace. The returned
color is in the RGBA basis.