Add MatterSim, Allegro and OCP models (fairchem-core) to ase_calculators

[janosh]

MatterSim does very well on the new thermal conductivity modeling task ($\kappa_{\rm{SRME}} = 0.554$, see MPA2suite/k_SRME#12, thanks @AntObi), almost as good as the current SOTA GRACE ($\kappa_{\rm{SRME}} = 0.525$, see https://matbench-discovery.materialsproject.org).

the small (1M) model interestingly does better than the large (5M) one, maybe a good sign that $\kappa_\rm{SRME}$ could prove more robust to overfitting

either way, MatterSim seems like a useful addition to the set of supported ASE calculators in atomate2, hence this PR.

adding Allegro and OCPCalculator as well since I've been using them with my local fork for a while

[gpetretto]

Sorry, I think it would be convenient to have these additional potentials among the easily accessible options. (maybe also including the latest UMA from Meta would be good). Is there any plan to merge this?

[JaGeo]

@gpetretto we can do this if someone can make sure all these potentials can be installed and tested (i.e. think of a good way to wrap those) or come up with an alternative suggestion of how to make sure the implementations keep working

[esoteric-ephemera]

@gpetretto and @JaGeo I'm in favor of including these - right now, we have dependency conflicts between mattersim and mace_torch that make it infeasible to test both in CI

In the longer term, I think we will need separate test blocks for different sets of MLFFs with compatible dependencies. Since we are not testing Allegro nor fairchem, I think it's also safe to avoid testing mattersim for now / skip any test if it's not installed (the default CI behavior)

[JaGeo]

@esoteric-ephemera fair for now.

Definitely, we need to come up with a testing strategy here in the future.

[esoteric-ephemera]

I'll post an issue for the missing tests after, I don't want to let it linger for too long

[JaGeo]

@esoteric-ephemera sure! And thank you!