Testing of the MOI wrapper

[odow]

I have many comments on the MOI wrapper, but there's a little hard to go through until the tests are passing. Fixing the tests should resolve most of my comments.

Functions like this strongly suggest that we're not testing the MOI wrapper correctly because it doesn't account for the different dual conventions between JuMP and Xpress:

XpressAPI.jl/ext/MOI_common.jl

Lines 909 to 930 in f20e039

	function MOI.get(model::Optimizer, result_index::MOI.ConstraintDual, indices::Vector{MOI.ConstraintIndex{F,S}})::Vector{Precision} where {
	F <: MOI.AbstractFunction, S <: MOI.AbstractSet
	}
	@assert result_index.result_index == 1
	return map(index -> begin
	c = _XPRS_constraint_from_moi_index(model, index)
	if c.xprs_index == -2 # Variable bounds are set as constraints with xprs index -2
	return NaN
	end
	# (so are parent constraints)
	if length(c.subconstraints) > 0 # Variable bounds are set as constraints with xprs index -2
	return map(subc -> NaN, c.subconstraints)
	end
	solstatus, solvec = XPRSgetduals(model.inner, c.solbuffer, c.xprs_index, c.xprs_index)
	if solstatus == XPRS_SOLAVAILABLE_NOTFOUND
	return NaN
	end
	return solvec[1]
	end,
	indices
	)
	end

In particular, we should be calling MOI.Test.runtests with minimal exclusions. Here's Xpress.jl:

https://github.com/jump-dev/Xpress.jl/blob/3f08ab203011da3375c1dd1af72cbb16c40b00bf/test/test_MOI_wrapper.jl#L47-L68

Here's how we handle duals in Xpress.jl:

https://github.com/jump-dev/Xpress.jl/blob/3f08ab203011da3375c1dd1af72cbb16c40b00bf/src/MOI/MOI_wrapper.jl#L3327-L3401

Note also our convention when there is a dual unbounded ray.

For our duality conventions, see https://jump.dev/MathOptInterface.jl/stable/background/duality/

For our infeasibility certificate conventions, see https://jump.dev/MathOptInterface.jl/stable/background/infeasibility_certificates/