Molgraph

Minor features and improvements
- `molgraph.applications.proteomics`
- Two different types of peptide models now exist --- one with, and one without, virtual/super nodes. For inclusion of super nodes specify `super_nodes=True` for `PeptideGraphEncoder`, otherwise `False`. Depending on `super_nodes` parameter, `PeptideModel` (aliased `PeptideGNN` or `PepGNN`) will return a Keras Sequential model with an certain readout layer.
- `molgraph.models.interpretability`
- Add `reduce_features` argument (default True) to `GradientActivationMapping`. Specifies whether node feature dimension should be averaged.

Major features and improvements
- `molgraph.applications.proteomics`
- `PeptideSaliency` is now based on the gradient (class) activation mapping algorithm. It considers all node features, including intermediate ones. Based on preliminary experiments, the saliencies looks more reasonable now.
- Users are now able to add their own dictionary of AA/Residue SMILES, see README at `molgraph/applications/proteomics/`.

Bug fixes
- `molgraph.applications.proteomics`
- Remove `keras.Sequential` wrapping around RNN and DNN of `PeptideGNN` to avoid 'graph disconnect' error.

Minor features and improvements
- `chemistry.MolecularGraphEncoder` are now by default *not* computnig positional encoding. Pass an `integer` to `positional_encoding_dim` to compute positional encodings of dim `integer`.

Bug fixes
- add `MANIFEST.in` and modify `setup.py` to include json files.

Bug fixes
- `molgraph.applications.proteomics`
- Fix import issue.