Add fallback for MOI.AbstractSymmetricMatrixSet{Triangle,Square}

[odow]

Closes #2093

@blegat is this what you had in mind?

It should allow something like

@variable(model, X[1:2, 1:2], Symmetric)
@constraint(model, X in SumOfSquares.ScaledDiagonallyDominantConeTriangle(2))

But what is a good model to test?

[codecov]

Codecov Report

Patch coverage: 100.00% and no project coverage change.

Comparison is base (ff69293) 97.97% compared to head (734d131) 97.97%.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #3424   +/-   ##
=======================================
  Coverage   97.97%   97.97%           
=======================================
  Files          34       34           
  Lines        4930     4939    +9     
=======================================
+ Hits         4830     4839    +9     
  Misses        100      100           

Impacted Files	Coverage Δ
src/macros.jl	98.63% <ø> (ø)
src/sd.jl	100.00% <100.00%> (ø)

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

[blegat]

I find it quite neat!
Once we have that, why no @constraint(model, (t, Symmetric(ones(2, 2))) in RootDetConeTriangle(2)) ?

[odow]

The main reason against (t, X) would be that we couldn't recover the function from the MOI model alone. We could add a new shape though.

@matbesancon suggested this a while ago: jump-dev/MathOptInterface.jl#1624

But it was in the context of MOI. It makes a bit more sense for JuMP. It'd simplify using RootDetConeTriangle if you didn't need to manually construct the triangular vector.

[odow]

I have a proof of concept for the tuple function. I'll merge this, then open a new PR.