So it's no longer necessary to provide the REFNAME environment variable.
Also the link template string format is now more general to support at
least gitlab github and cgit.

'box-forced' is deprecated as of matplotlib 2.2.

Previously, in order to not to fix the matplotlib backend, we required
users to import matplotlib.pyplot before calling any Kwant plotting
functions.  This did not have the desired effect, since we did import
`matplotlib.backends` and that also fixes the backend.

Now, both backends and pyplot are imported at the last possible moment
and a warning is emitted if this fixes the backend.

The problem was introduced in commit 7c802539.

Previously the fact that the positive direction of mode
velocity is determined by the translational symmetry vector
of the system was not made clear in the relevant docstrings.

Closes #143.

The lattice constant is never changed from 1, and in addition can be
varied independently of the 't' and 'α' parameters; This should not
be the case if the model is meant to be derived from a discretization
of a continuum model. We remove the lattice constant to limit confusion.

For a spin-orbit term of the form '-iασ_y∂_x' the hopping
for 'HoppingKind((1, 0), lat)' (*from* 'x' *to* 'x + a')
is 'iασ_y/2'.

Closes #175.

Previously, an erroneous current was shown at *zero* magnetic field,
which was due to a choice of degenerate eigenvalue. Now we evaluate
the current and density at finite magnetic field, to demonstrate
the intended effect.

Closes #150.

Produce vector graphics, rather than rasterizing,
as rsvg-convert does.

Add the 'unicode-math' package and set the
latex engine to 'xelatex'.

Closes #126.

Closes #166.

Previously a point defect introduced by an incorrect treatment
of the origin produced erroneous scattering. We also update the text
to reflect the actual magnetic potential used.

Closes #156.

I assume that most people encounter the tutorial example scripts by
reading the documentation, and not by viewing the tutorial subdirectory
of a checked-out Kwant source.

That's the motivation for moving all the manipulations of tutorial
scripts from setup.py to doc/Makefile.  Previously, a successful 'make
html' would require a preceding execution of 'setup.py build_tut'.  Now,
a simple 'make html' is enough.

While at it, I reorganized where the example scripts and their outputs
are stored.  Everything is now in 'doc/source/code' with its three
subdirectories 'download' (to be shown to readers), 'include' (with
include markers), and 'figure' (figure generation & figures).  This
organization is clearer and also makes the generation of figures
separate from the tutorial.

The basic idea remains the same, but should work much better now.  The
difference is that images/example.py.diff is now a patch with complete
context and becomes the authoritative source for both the visible flavor
of an example as well as for its figure-generating variant.  Both
tutorial/example.py and images/example.py are extracted from this file
by 'make html'.  Thanks to the complete context the diffs are quite
readable and may be modified directly.

Alternatively, one may also modify the generated scripts.  When
tutorial/example.py has been modified, it will be patched and saved as
images/example.py.  The patching is done using the tool 'wiggle' that
works much better than 'patch'.  If a conflict occurs, conflict markers
are added to the output file and its dated back to the dawn of time
(i.e. 1970) in order to mark the conflicts as not yet resolved.  After
resolving, 'make html' is simply run again.

Upon a successful execution of the figure-generating script the diff
gets recreated automatically, 'generate-diffs' no longer exists.

These classes used not to have any features beyond
kwant.system.{Finite,Infinite}System, but this is no longer the case.

lowercase FAQ.rst, FAQ.py, etc., give meaningful names to labels,
and stop building an user-visible tutorial/faq.py

Apply pep8 fixes and refactor some sections to be more idiomatic
(e.g. avoiding triple nested loops). Also some minor reorganisation
of the sections to give the text a better flow.

 The total density of states now integrates to `N`, the size of the system.

In addition to basic updates to skip permanent redirect, remove an
obsolete reference to macports installation instructions on the main
website.

  `integrate` is a more accurate name for a function that integrates
   the spectral density. Docstrings and tutorial updated.

closes #125