Fix performance issue building large sparse linear operators

[odow]

Closes #121

I think I had just assumed that people would build lots of small rows, rather than a big single complementarity constraint.

For n = 2048 I now get:

julia> @constraint(model, M * x .+ q ⟂ x)
[x[1] - 1, 2 x[1] + x[2] - 1, 2 x[1] + 2 x[2] + x[3] - 1, 2 x[1] + 2 x[2] + 2 x[3] + x[4] - 1, 2 x[1] + 2 x[2] + 2 x[3] + 2 x[4] + x[5] - 1, 2 x[1] + 2 x[2] + 2 x[3] + 2 x[4] + 2 x[5] + x[6] - 1, 2 x[1] + 2 x[2] + 2 x[3] + 2 x[4] + 2 x[5] + 2 x[6] + x[7] - 1, 2 x[1] + 2 x[2] + 2 x[3] + 2 x[4] + 2 x[5] + 2 x[6] + 2 x[7] + x[8] - 1, 2 x[1] + 2 x[2] + 2 x[3] + 2 x[4] + 2 x[5] + 2 x[6] + 2 x[7] + 2 x[8] + x[9] - 1, 2 x[1] + 2 x[2] + 2 x[3] + 2 x[4] + 2 x[5] + 2 x[6] + 2 x[7] + 2 x[8] + 2 x[9] + x[10] - 1, 2 x[1] + 2 x[2] + 2 x[3] + 2 x[4] + 2 x[5] + 2 x[6] + 2 x[7] + 2 x[8] + 2 x[9] + 2 x[10] + x[11] - 1, 2 x[1] + 2 x[2] + 2 x[3] + 2 x[4] + 2 x[5] + 2 x[6] + 2 x[7] + 2 x[8] + 2 x[9] + 2 x[10] + 2 x[11] + x[12] - 1, 2 x[1] + 2 x[2] + 2 x[3] + 2 x[4] + 2 x[5] + 2 x[6] + 2 x[7] + 2 x[8] + 2 x[9] + 2 x[10] + 2 x[11] + 2 x[12] + x[13] - 1, 2 x[1] + 2 x[2] + 2 x[3] + 2 x[4] + 2 x[5] + 2 x[6] + 2 x[7] + 2 x[8] + 2 x[9] + 2 x[10] + 2 x[11] + 2 x[12] + 2 x[13] + x[14] - 1, 2 x[1] + 2 x[2] + 2 x[3] + 2 x[4] + 2 x[5] + 2 x[6] + 2 x[7] + 2 x[8] + 2 x[9] + 2 x[10] + 2 x[11] + 2 x[12] + 2 x[13] + 2 x[14] + x[15] - 1, [[...4066 terms omitted...]], x[2034], x[2035], x[2036], x[2037], x[2038], x[2039], x[2040], x[2041], x[2042], x[2043], x[2044], x[2045], x[2046], x[2047], x[2048]] ∈ MathOptInterface.Complements(4096)

julia> @time optimize!(model)
  0.499948 seconds (26.03 k allocations: 297.203 MiB, 46.23% gc time)