per-atom PAE scores

[aozalevsky]

models.qa_metric.LocalPairwise is defined on the per-residue level, but as @arogozhnikov explained in chaidiscovery/chai-lab#52, they estimate per-atom PAEs for ligands and non-canonical amino acids/nucleotides (atom vs residue (for canonical polymers) and atom vs atom for ligands and non-canonical residues). It would be nice if we could extend the qa_metric.LocalPairwise definition.

[gtauriello]

I created a new issue at #21 which should cover this case additionally to all other quality estimates introduced with AF3. If you think that #21 captures everything needed for chaidiscovery/chai-lab#52 I think we can close this one as redundant?

[aozalevsky]

sure