Add the sparse SDP relaxation

[ivanightingale]

We have implemented an equivalent of PowerModels.SparseSDPWRMPowerModel formulation in PGLearn.jl. This is a feature request to add the formulation to this codebase.

[ccoffrin]

I am open to a PR that proposes that formulation in a similar format to the models in scr/model/opf.jl.

I do worry a bit about the numerical stability of this approach and that long term maintenance could be difficult.

[ivanightingale]

@ccoffrin Could you be more specific about what you refer to as "this approach"? Do you mean the sprase SDP relaxation itself or something implementation-wise? I'm confident that the implementation in PGLearn.jl matches that of PowerModels.SparseSDPWRMPowerModel mathematically (need to verify with unit tests).

It's straightforward to slightly modify the PGLearn.jl code to match the format of the models in scr/model/opf.jl. A caveat is it might be necessary to use/copy helper functions in PowerModels.jl related to clique decomposition:
https://github.com/lanl-ansi/PowerModels.jl/blob/bcacabf5d931ea7d8725363c3cc3d3eb91803d04/src/form/wrm.jl#L141-L147