Distributed section needs improvements

[carstenbauer]

• Last time I checked SharedArray only worked when shared memory was available, i.e., when all workers run on a single machine. If that's still the case, then it is not generally useful when doing "real" distributed computing involving multiple machines.
• IMO the most important help that one can provide for Distributed.jl is to tell people how to practically start workers on multiple machines. Keyword arguments for addprocs like exename or dir are important if the file system isn't shared (Julia and the working directory might be at a different location than on the main computer). On cluster, you will likely need to use ClusterManagers.jl or learn how to get the hostnames of the machines from a job scheduler (like SLURM). OTOH, you might want to use MPI here anyways.
• One could showcase an array-abstraction for Distributed like DistributedArrays.jl's or Dagger.jl's DArray (similar to CuArray for GPUs).
• ...

If desired, I think I might be able to help with improving the section.

[gdalle]

Thanks, some help would be lovely here! @jacobusmmsmit wrote this section last week and I did some shallow editing to get it ready for JuliaCon but I didn't check the whole content cause I never used distributed computing. Excited to a review a PR in this direction