add MOI interface for complex PSD cone

[araujoms]

I've added a MOI interface for the complex PSD cone according to the design @blegat suggested here, adding the sets ScaledComplexPSDCone and ComplexPositiveSemidefiniteConeTriangle and using the chain of bridges
MOI.HermitianPositiveSemidefiniteConeTriangle -> ComplexPositiveSemidefiniteConeTriangle -> MOI.Scaled{ComplexPositiveSemidefiniteConeTriangle} -> ScaledComplexPSDCone.

Still missing is the middle part ComplexPositiveSemidefiniteConeTriangle -> MOI.Scaled{ComplexPositiveSemidefiniteConeTriangle}, which I don't know how to do, but I expect it will be rather simple. I added the functions set_dot and set_coefficients for ComplexPositiveSemidefiniteConeTriangle, which I assume are necessary for MOI.Scaled to work.

[araujoms]

Now everything is working, except that JuMP doesn't want to use the new cone! I have to explicitly remove the inefficient bridge via remove_bridge(model, MOI.Bridges.Constraint.HermitianToSymmetricPSDBridge) for it to use the complex cone.

[blegat]

You can tweak the bridging cost to make that your bridges are preferred over the inefficient one.
Use something like

MOI.Bridges.bridging_cost(::Type{<:YourBridge}) = 0.1

[araujoms]

That was it, thanks a lot.

[blegat]

Looks good to me, the buildkit failure is unrelated

[kalmarek]

@araujoms the only question that comes to my mind is: Is the interface/bridge really covered by the tests? I can only see a set of tests removed in this PR :D

[araujoms]

That was a list of the MOI tests to be excluded, so by removing it I'm actually adding it =)

But the honest answer to your question is that I don't know. I assume MOI has plenty of tests already for the complex PSD cone. If that's not the case I'm happy to write one.

I did add a test of the low-level interface in a previous PR.

[blegat]

Now that the bridging costs are lowered, it should be used by the MOI tests.

[odow]

I'm happy if @blegat is happy