Adjust manifold vector forms to use space dimension

Closes #55

See merge request !50

Overhaul of form generator

See merge request !49

Generation of theses forms did no longer work due to problems with the update
from a scalar to a vectorial treatment of trial and test spaces.

The commit fixes this, by making form_func() convert trial and test to
TensorialVectorFunctionSpace internally.

Note: I am not too happy with the changes needed to statisfy mypy.
      I was not able to keep the differentiation between TestSpace
      and TrialSpace for the arguments that are of type
      TensorialVectorFunctionSpace. Ideas welcome.

When --geometry was not given we used args.geometry, but should
have used form_info.supported_geometry_options instead.

Rationale: The epsiloncc forms cannot be generated anyway.
Trying to do so raise an exception:
HOGException("Constant viscosity currently not supported.")

The commit changes the behaviour of generate_all_hyteg_forms.py with respect
to the selection of forms to generate. It introduces a new function

   assemble_list_of_forms_to_generate()

that extracts the list of forms to generate from the list of defined forms
based on the command-line options in the following fashion:

* if neither --geometry, nor --filter is given, all defined forms are
  selected

* if only --filter is given, it is applied to the list of all defined forms,
  which are generated for each geometry they support

* if only --geometry is given, it extracts all forms that support the given
  geometry element type

* if both --filter and --geometry are given, the filter gets applied first;
  then the function removes all froms that do not support the selected
  geometry

The commit touches on MR !48 and issue #45.

As a spin-off the commit makes the FormInfo class sortable.

The attribute is a list with string elements from the set "triangle",
"tetrahedron", "embedded_triangle". This allows to describe the element
geometries for which the form is intended.

The choices for "--geometry" before this commit are

["triangle", "tetrahedron", "embedded_triangle", "both"]

with "both" being the default. This commit changes them to

["triangle", "tetrahedron", "triangle+tetrahedron", "embedded_triangle"]

with "triangle+tetrahedron" being the default.

Divergence and Gradient Forms for Conforming Crouzeix-Raviart element in 2D

See merge request !47

The level of nesting in the if-statements apparently was too much
for mypy to understand.

Notes:
- The functionality is required for issue #49.
- Generation of DG-DG operators currently seems to be a more theoretical
  perspective. See issue #19.

Status:

- The high level part, such as form definition and name generation works.
- We can successfully generate fro triangles the following forms:
  * dg1_to_p2_plus_bubble_divt_affine_q3
  * dg1_to_p2_plus_bubble_divt_blending_q3
  * p2_plus_bubble_to_dg1_div_affine_q3
  * p2_plus_bubble_to_dg1_div_blending_q3

The commit adds a DGFunctionSpace to represent discontinuous piecewise
polynomial functions a.k.a. Discontinuous Galerkin. Currently it supports
degrees 1 and 2 in 2D and 3D.

Update location of header P2PlusBubbleFunction.hpp in HyTeG

See merge request !46

Following MR hyteg!887 the header file P2PlusBubbleFunction.hpp
no longer resides in experimental, but in p2functionspace.

Clean up indexing, Pt. 1

Closes #46

See merge request !43

Give hint to user on the need to extend generate.py

Closes #53

See merge request !45

Fixes #53

Fix pyproject.toml for wheel generation

See merge request !44

The packages list was missing the new module introduced in !37 [c3815303].
Commit adds hog.operator_generation.function_space_implementations to list.
This is a recurring problem, see also !22 in this respect.

Resolve issue 48

Closes #48

See merge request !42

This should temporarily fix the problem in the "form-generation" job of the
pipeline with the p2_plus_bubble_diffusion_affine_qe form. For the moment we
place it in the p2 folder. However, file_path() needs refactoring, especially
if we add more spaces.

The commit
- adds component_index as keyword to IntegrandSymbols
- makes gradient and divergence form pass component_index to process_integrand
- the corresponding integrands now check the shape of grad_u resp. grad_v
  and, if it is not square, revert to extracting the corresponding component
- moves the is_implemented check from the call of HyTeGIntegrator outwards,
  so that the call to form_func is protected against out-of-bounds indices

Add support for bubble enhanced P2 element in 2D

See merge request !40

Make LagrangianFunctionSpace.is_continuous return False for P0

See merge request !41

Note that we return False for is_affine(), because there seems to be a problem
with the implementation of the optimisation based on this for True.

Note that this commit includes an IMHO rather not so nice hack for
the facedof_index function. The latter assumes that index is
Tuple[int, int, int]. However, the index it receives contains only
two entries.

Note that the ordering of the shape functions associated with the edge dofs
needs to be the one used by FEniCS/HyTeG.

Note that this is only for testing form generation and verification of the
result. The generated forms are, currently, not useable in HyTeG as there
is neither a base class for them nor any operator that could make use of
them.