Add fallback for ListOfConstraintIndices and NumberOfConstraints

[odow]

This PR adds a fallback for ListOfConstraintIndices and NumberOfConstraints to return default values in the case that the solver does't support the particular function-set combination.

This has come up a few times:

Closes #2011
Closes #2010
Follow-up to #2008

We already had something similar in two places:

MathOptInterface.jl/src/Utilities/struct_of_constraints.jl

Lines 142 to 150 in 71defad

	function MOI.get(
	model::StructOfConstraints,
	attr::MOI.ListOfConstraintIndices{F,S},
	) where {F,S}
	if !MOI.supports_constraint(model, F, S)
	return MOI.ConstraintIndex{F,S}[]
	end
	return MOI.get(constraints(model, F, S), attr)
	end

MathOptInterface.jl/src/Utilities/struct_of_constraints.jl

Lines 132 to 140 in 71defad

	function MOI.get(
	model::StructOfConstraints,
	attr::MOI.NumberOfConstraints{F,S},
	)::Int64 where {F,S}
	if !MOI.supports_constraint(model, F, S)
	return 0
	end
	return MOI.get(constraints(model, F, S), attr)
	end

#2008 added it to two more:

MathOptInterface.jl/src/Utilities/model.jl

Lines 416 to 424 in 71defad

	function MOI.get(
	model::AbstractModel,
	attr::MOI.NumberOfConstraints{MOI.VariableIndex,S},
	)::Int64 where {S}
	if !MOI.supports_constraint(model, MOI.VariableIndex, S)
	return 0
	end
	return MOI.get(model.variables, attr)
	end

MathOptInterface.jl/src/Utilities/model.jl

Lines 443 to 451 in 71defad

	function MOI.get(
	model::AbstractModel,
	attr::MOI.ListOfConstraintIndices{MOI.VariableIndex,S},
	) where {S}
	if !MOI.supports_constraint(model, MOI.VariableIndex, S)
	return MOI.ConstraintIndex{MOI.VariableIndex,S}[]
	end
	return MOI.get(model.variables, attr)
	end

So this PR is essentially just making some implied behavior concrete.

I've tested with Gurobi:

julia> model = Gurobi.Optimizer();

julia> MOI.Test.test_attribute_unsupported_list_of_constraint_indices(model, config)

But once this PR is open, I'll run SolverTests to see the damage: jump-dev/SolverTests#19

[odow]

No issues identified by SolverTests, jump-dev/SolverTests#19, so I'd say this is okay.

[blegat]

Since all MOI wrappers tested by SolverTests are correctly implemented, this is not different from #2011.
However, as detailed in #2011 (comment), I don't see in which situation, it is a good thing to add a fallback for supported constraint types.
If a solver supports a constraint type and does not implement the getter, returning zero is never a good implementation, a clear error saying that this is not implemented by the MOI implementation so a cache should be used if that feature is needed is what is most suited in that situation

[odow]

Added a GetAttributeNotAllowedError

[blegat]

supports_constraint is not enough, consider for instance CSDP with VectorOfVariables in PSD, we need some condition like in #2011

[odow]

Done.

[blegat]

Looks good, maybe add a comment in the code since it is not so obvious why we are doing what we are doing ?