Tests: add partial_name_map_composite fixture

- the partial_name_map_composites fixture creates a random (but seeded)
  partial name_map by the actual associations (i.e. one name to one
  rule)
- this is used to correctly test MixedComposites/NameLayerMapComposites
  to fall back to the LayerMap
- add pyrng fixture, based on the same seed as the numpy RNG
- removed unneeded input fixtures for some fixtures in conftest.py

Docs: Better image classification tutorial

- change the VGG tutorial to an image classification tutorial using
  VGG11 (with and without batch-norm) and ResNet18
- demonstrate Saliency Map and SmoothGrad first
- demonstrate EpsilonGammaBox and EpsilonPlusFlat
- demonstrate how to modify the stabilizer and epsilon of the built-in
  composites
- demonstrate different visualizations
- demonstrate how to modify the built-in composites using the
  rule-parameters and cooperative layer maps
- demonstrate a custom NameMapComposite
- demonstrate a custom LayerMapComposite
- add a ipynb-based table of contents for easier navigation and an
  overview
- change deprecated torchvision model argument `pretrained=False` to
  `weights=None` and `pretrained=True` to `weights='IMAGENET1K_V1'`

Composites: Add epsilon and gamma parameters

- add epsilon parameter to EpsilonGammaBox, EpsilonPlus, EpsilonAlpha2Beta1, EpsilonPlusFlat, EpsilonALpha2Beta1Flat
- add gamma parameter to EpsilonGammaBox

README: Update, better usage, relative links

In README.md:
- bring the Usage part up-to-date, with some better formatting
- add a small demonstration in Usage
- move Documentation to its own Section, so it can be found more easily
- remove statement about "interfaces can break anytime", the interface
  should be mostly stable now
- add another link to the documentation after Usage
- give more details in the introduction to the Example
- add a section Example Heatmaps, which shows example heatmaps for
  various supported attribution methods for VGG16 and ResNet50 produced
  by feed_forward.py
- add a sub-section about the documentation in Contributing with notes
  about building
- make absolute github links relative, such that they can point to their
  respective revision. They were previously absolute links for display
  on PyPI, which cannot correctly disply relative files.
- add badge for tests
- fix inconsistent capitalizations
- fix Pytest link pointing to Pylint

New images:
- add two images to show heatmaps for VGG16 and ResNet50

In setup.py
- define a function to fetch the long description for PyPI, which
  replaces relative links with absolute ones on github. This fixes the
  previous problem with relative links on PyPI, and will also display
  and reference files for the correct release, instead of always for
  master.
- due to `python -m build` by default copying all files for the sdist,
  and then building the wheel in the same folder in an isolated
  environment, the revison would not be correct within the wheel. The
  sdist however creates a PKG-INFO file, which will contain the
  long_description with the absolute links, which will be used if the
  git version cannot be used. The fallback is to use master, as done
  previously.

Test Simple Layer Rules

- added simple module fixtures, for simple, element-wise, parameter-less
  modules like ReLU
- added simple data fixtures for the simple layers
- made batch-size a pytest CLI option, defaulting to 4
- added tests for rules that only apply to simple layers

How-To for Visualization of Results

- added an elaborate how-to on visualizing results
- added a subsection about heatmap normalization
- added a subsection about color maps, which shows built-in color maps
  and how to use them
- added a subsection on the Color-Map Specification Language (CMSL),
  with a simple grammar specification, a detailed explanation of its
  workings and a few examples along visualizations
- added a subsection on changing palettes
- added compile-time visualizations of heatmaps using
  sphinxcontrib.datatemplates
- added sphinx.datatemplates requirement
- added config option `generated_path` to choose where generated images
  should be stored

Added more badges

- PyPI version and license

Added .readthedocs.yaml file

- removed zennit requirement from docs

Fix the Norm and bias of AlphaBeta

- the normalizing terms of AlphaBeta were implemented wrong, where the
  parts of the positive and negative terms respectively were split,
  which was fixed
- the negative bias contributed twice in AlphaBeta

Added a few new built-in colormaps

- france, which goes from blue to white to red, in the exact colors of the french flag
- seismic, which is blue to white to red with brightness d0
- coolio, which is cyan to white to magenta with brightness d0
- coleus, which is green to white to magenta with brightness d0