Logic and subvariable substitution with bounds

[aescande]

Right now, we are sometimes conservative on what constraints can be viewed as bounds.
Ideally, we would like any constraints that can be rewritten as l <= x <= u to qualify as such (this includes constraints such as l <= Dx <= u, with D diagonal and invertible).
Some constraints might be bounds only if the solver honors a substitution hint, for example the constraints l <= x + y <= u with y substituted from y = D x. We should see what to do with those. For now, the logic makes it so that a constraint with 2+ variables can never be seen has a bound.

On the topic of substitution, some subvariable substitutions, such as those found in mimic joints, would retain the bound qualification of a constraint, but might split the variable in two. For now, we don't have the logic and pipeline to handle this case, and we conservatively discard this case (see 5f44ec1). A venue might be to split the constraints into two in this case, but with the mean to associate the original one with the two new ones. Another possibility would be to somehow accept bounds with several variables.