Add assembleGradient() method to FE assemblers

See merge request !72

This adds a method to compute only the gradient.

On the pro side it saves
a lot of time if only the gradient is needed (for first order methods).

On the con side it introduces some duplicated code (but only for hyberdual
numbers).

A test case is added.

Some improvements on exception handling

See merge request agnumpde/dune-elasticity!74

Otherwise the type of the exception gets lost.

Some regularizing algorithms based on exceptions filter the exceptions by type to detect math errors such as ill-formed systems or inputs.

Rename deprecated getCollectiveCommunication()

See merge request !73

See https://gitlab.dune-project.org/core/dune-common/-/merge_requests/1091

ADOL-C: do not put the rank into trace_off()

See merge request !70

Following the doc at

https://github.com/coin-or/ADOL-C/blob/master/ADOL-C/doc/adolc-manual.pdf

this enforces writing the tapes to disk and has nothing to do with the actual
taping. From the doc:

"By setting the optional integer argument file of trace off to 1, the user may force a
numbered tape file to be written even if the tape array (buffer) does not overflow. If the
argument file is omitted, it defaults to 0, so that the tape array is written onto a tape file
only if the length of any of the buffers exceeds [OLVT]BUFSIZE elements."

Fix in Mooney-Rivlin density: Use proper 4th term for Ciarlet

See merge request !69

In Ciarlet the last factor is left unspecified. I could not follow the
choice used in this implementation. I guess it was intended to minimize
the energy under a hydrostatic stress F = t*I at the unit matrix, i.e.,
for t=1. A simple calculation shows that (independent on the dimension)

d = 2a + 4b + 2c

has to hold in this case. Values in the test were adjusted.

Improve hyperdual class

See merge request !68

Fix in mooneyrivlindensity

See merge request !67

Square the determinant in the Ciarlet variant

Replace DUNE_DEPRECATED by [[deprecated]]

See merge request !65

It depends on Amiramesh which is dropped in dune-grid since 2.8

Cmake: Look for PYTHONLIBS and for Python3

See merge request !62

This basically reverts f10d311d

Fix build and execution of finite-strain-elasticity

See merge request !61

[CI] Also test against Dune 2.7

See merge request !60

Because in this particular combination, no cmake Python test is ever
called, and therefore Python is not available.

Even for the master branch I'd like to know that this kinda works
against the 2.7 release, because that is what people using
packages have.

Some small modernizations

See merge request !59

There already was code that printed a warning when the trust-region
radius dropped below the required solver tolerance.  (If that happens,
the solver terminates, because the small trust-region forces the
correction to be smaller than the required tolerance.  However,
the code didn't warn the user about this, and such cases almost
looked like successful runs.)

Unfortunately, the warning code didn't work, because the standard
solver termination criterion stopped the TR iteration before
reaching the check for the trust-region size.

Make FEAssembler store a const reference of the basis

See merge request !57