Don't add extra slacks (artificial variables) for ranged rows

[chris-maes]

When we have a ranged row of the form

l_i <= a_i^T x <= u_i

we convert it to an equality constraint for dual simplex by adding a slack s_i

a_i^T x - s_i = 0, l_i <= s_i <= u_i

We were also adding an additional slack variable for every equality constraint to ensure A was full row rank and we had a slack basis. For ranged rows this was leading to

a_i^T x - s_i + s_k = 0, l_i <= s_i <= u_i, 0 <= s_k <= 0.

This PR makes it so we no longer add s_k to rows that came from ranged rows.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.