Integrate nmer into mhc_epitope to allow packer-compatibility
The "established" mechanism for de-immunizing proteins in Rosetta is using the NMerSVMEnergy class implemented by @indigogo . In contrast to MHCEpitopeEnergy, NMer is not inherently packer compatible.
This PR essentially adds a new MHCEpitopePredictor that allows nmer to be packer-compatible, using the framework we established with mhc_epitope. While the SVM-based scoring is rather slow using the current SVMs, the common framework could still be used to score proteins outside of the context of the packer. In addition, if faster SVMs are added (or generated by the user), they could be used in a packer-compatible manner.
In order to implement this, a few changes to both the MHCEpitopeEnergy class and the NMer classes needed to be made. Testing has not shown any changes triggered by these changes. (The failure for mhc_epitope_nmer_preload is expected, as the test was changed to introduce nmer stuff. The failure for mhc_epitope is cosmetic.)
In addition to the nmer integration itself, this PR also makes some under-the-hood changes:
-MHCEpitopeEnergy now handles Predictors that work using a core + potentially missing overhang definition of their peptides, where the "overhang" regions could hang off the end of the peptide chain. This is different from cases where the entire peptide is "core," meaning that some peptides at the end of the chain would not be counted. NetMHCII dealt with this internally, but nmer does not, so it was needed for equivalent treatment. Extendable to other Predictors that use this strategy.
-Disk reads are now all handled by the ScoringManager, which should improve performance in certain cases.
-Some things got moved around to address the split of `core.3` which took place some months ago.
Thanks to @indigogo for originally writing the nmer code, and @vmullig for review (and particularly suggesting the ScoringManager improvements).