GeodesicFEAssembler can replace it without problems.

Capturing any polymorphic object by value is wrong,
and gcc 8 started to warn about this in the case
of exceptions.

This fixes #1

The formula for the FD gradient never took the distance between
the two Lagrange points into account, even though the gradient
of an interpolation function clearly depends on it.

(cherry picked from commit 7f326ded)

Needed for projection-based finite elements.

This patch also adds a unit test.  So far this test passes.

[[Imported from SVN: r8348]]

[[Imported from SVN: r8335]]

[[Imported from SVN: r8334]]

[[Imported from SVN: r8204]]

[[Imported from SVN: r8194]]

[[Imported from SVN: r8193]]

[[Imported from SVN: r8192]]

[[Imported from SVN: r5940]]

[[Imported from SVN: r2010]]