Add docs to caltech cluster

[simone-silvestri]

tries to fix #3779

This PR also switches the parallelization of the docs from Distributed to MPI, with a couple of advantages:

• we can use more CPUs to run the docs which should speed up the computation
• we can run distributed Oceananigans example in the docs, both on CPUs and GPUs

This last new feature probably warrants (in another PR) writing a distributed example

[simone-silvestri]

we probably need to add secret tokens and deploy docs ssh keys to the cluster to make this work

[glwagner]

I'm skeptical about this because our docs are super computationally intensive.

[simone-silvestri]

All the more reason to use a better node rather than Tartarus

[glwagner]

Tartarus is probably a lot faster than the caltech nodes?

[simone-silvestri]

We decided not to do this.