Add support for dense models and GNINA default model

[RMeli]

Description

Thanks to @drewnutt #43 , the correct weights for the dense models are now available, which allow to support the dense models alongside default2017 and default2018. More importantly, this allows to build the default model ensemble, thus reproducing all results obtained with GNINA.

PR Checklist

• Tests
• Documentation

[codecov]

Codecov Report

Merging #44 (57c82a5) into main (1b768a9) will increase coverage by 0.01%.
The diff coverage is 100.00%.

[RMeli]

In the GNINA 1.0 paper there is a small typo: the default ensemble uses the redock_default2018_2 model, instead of the redock_default2018 model (as written in the paper).

GNINA 1.0 implementation, for reference:

https://github.com/gnina/gnina/blob/aab122edf91b59c07de48a0922256f34a809c49b/gninasrc/lib/cnn_scorer.cpp#L102-L109