re-introduce gdock module

[rvhonorato]

Checklist

• Tests added for the new code
• Documentation added for the code changes
• Modifications / enhancements are reflected on the haddock3 user-manual
  • non-applicable, we still need to do some deeper evaluation of the integration
• CHANGELOG.md is updated to incorporate new changes
• Does not break licensing
  • gdock has a zero-clause license so it's fully compatible here
• Does not add any dependencies, if it does please add a thorough explanation
  • adds gdock-py which has no parent dependencies

Summary of the Pull Request

This PR re-introduces gdock as a sampling module. It was first introduced in #72 and then removed in #1454. The reason for removal is that gdock depended on an external dependency for its scoring function that was no longer available - so the whole software was not functional anymore.

Since then I have completely re-written gdock from scratch and implemented the energy calculation myself (see the release notes for more info about the re-write)

To fulfill this integration into haddock3 I have written thin bindings (bits of code that wrap the gdock's source code in python) to make the addition here simpler - see gdock-py.

Since haddock uses ambiguous restraints, I have also added this concept in gdock so it can seamlessly read haddock's tbl file and can handle the OR logic. I have implemented, to the best of my knowledge, the same function that haddock uses for its AIR restraint calculation - https://github.com/rvhonorato/gdock/pull/64

As for the integration itself I also added some compatibility to be in line with the other sampling module; [gdock] can handle ensembles by running each combination and it can also handle crossdock=True.

The key parameters I have exposed in the gdock interface are:

Parameter	Default	Description
ambig_fname	""	AIR restraints .tbl file
max_generations	250	GA generation limit
number_of_individuals	150	GA population size
sampling	1000	Models to collect per input pair**
seed	42	Random seed
crossdock	true	Dock all receptor × ligand combinations

**sampling is a special parameter I added just for this integration, it will go bypass gdock's selection and return up to N models that are different enough based on its HoF (Hall-of-Fame) definition

I have also created some example files for the compatible scenarios replacing rigidbody with gdock - however we cannot make a one-to-one comparison given that gdock and haddock are using different formulations of the OPLS forcefields.

We would need a deeper investigation to properly parametrize which parameter and module configurations we should use to obtain similar results - this is however way far beyond the scope of this PR.

Here I am "just" adding gdock as a module and I think this can also serve as a good example of how to add new (external)-modules, so please keep the discussion in this PR so it serves for future reference.

Related Issue

#1595

Additional Info

Please have a look at gdock.org for full information about how gdock works. I did not add the whole information here since its described elsewhere.

[rvhonorato]

The gdock module is properly implemented and functional.

However I've hit a conceptual wall in the way restraints are treated. gdock currently (v2.1.0) can only handle PAIRS of restraints. So while running a gdock x rigidbody comparison in the cherry-picked examples I was naively just "reducing" the restraints to a set of pairs, but this creates a set of impossible to satisfy constraints which results in models with a lot of clashes.

So I will pause work on this while I implement ambiguous restraints in gdock, follow here: https://github.com/rvhonorato/gdock/pull/64

For now I'll keep the PR open since its just additive - but if the code diverges too much I'll reopen it.

[rvhonorato]

@amjjbonvin please add whoever else you think is relevant for this review