Dependency fixes

Closes #386

See merge request kwant/kwant!370

This is a temporary choice due to a bug in IPykernel
https://github.com/ipython/ipykernel/issues/540

Closes #386

We plan a release no earlier than September 2020, which means that
we target Debian buster and Ubuntu focal when determining package
versions

Fix reflection calculation for kwant.greens_function

Closes #398

See merge request kwant/kwant!377

We are typically concerned with features no smaller than 1e-4 - 1e-5
times the bandwidth. With this change we add some tolerance to
error accumulation while not influencing physical results.

The broadening matrix (gamma) itself has norm ~0 when there are no
open modes, meaning that the tolerance we are checking against is
always too small in this case. Instead we check against the norm
of the selfenergy, which is nonzero even when there are no open modes.

A global constant named 'chain' is defined later in the test module,
so 'from itertools import chain' is confusing.

Describe more fundamental attributes first.

Builder.substituted doesn't change the builder in place, contrary to
what the docstring suggested.

fix eradicate dangling across translational symmetry

See merge request kwant/kwant!369

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'