Handle dual and reduced cost

[tbittar]

Summary

This PR adds support for dual variables (constraint shadow prices) and reduced costs (variable marginal costs) as first-class model outputs in GemsPy.

New expression operators

Two new operators are available in model expression strings:

• dual(constraint_name) — returns the shadow price of a constraint at each time step
• reduced_cost(variable_name) — returns the reduced cost of a variable at each time step

Both are parsed and validated at model-build time via the visitor pattern (DualNode, ReducedCostNode). ModelIdentifiers now carries the set of constraint names so that dual() arguments can be checked against known constraints.

Simulation table output

SimulationTableBuilder collects and exposes duals and reduced costs alongside other extra outputs:

• Constraint duals are read from linopy's model.dual dataset; the builder aggregates __eq, __lb, and __ub suffixes into a single value per constraint.
• Reduced costs require solver-specific extraction because linopy does not expose a unified API for them (see below).

Solver-specific reduced cost support

Linopy provides no generic reduced-cost accessor, so the builder falls back to each solver's native API:

Solver	Method	Notes
Xpress >= 9.8	solver_model.getLpSol()	Returns (x, slack, duals, djs); DJ list used. Checked before getSolution because xpress.problem also exposes getSolution with an incompatible signature.
Gurobi	solver_model.getAttr("RC", solver_model.getVars())	Standard Gurobi attribute query.
HiGHS	solution.col_dual	Column dual from HiGHS solution object.

If the solver model is None or unsupported, reduced costs are silently skipped (returns {}).

Project environment

• A new solvers optional dependency group (xpress>=9.8, gurobipy>=10.0) is added to pyproject.toml for running solver-specific tests locally with a free licence.
• CI installs this group (uv sync --group dev --group solvers) so solver-specific tests run in the pipeline.

Pre-commit consistency

• Removed the redundant --config pyproject.toml argument from the black hook (black discovers pyproject.toml automatically).
• Simplified the isort hook args to --profile black (removed explicit --filter-files src tests which conflicted with pre-commit's own file filtering).

Test plan

• pytest tests/unittests/ — unit tests covering new expression nodes, simulation table builder, and fluent accessor API
• pytest tests/e2e/functional/test_dual_reduced_cost.py — end-to-end tests for studies 10_5, 10_5_1, 10_5_2 (requires solver licence)
• pre-commit run --all-files — confirm black and isort hooks pass cleanly

Closes #182

🤖 Generated with Claude Code

[aoustry]

Shall we update the docstring?

[aoustry]

Not sure to understand why it is forbidden. The purpose of authorizing dual/rc in port-field is indeed to receive them in other ports?

Once again : I would say that we can just decide on the degree of the expressions (variable bounds, constraint, binding-constraint or objective). This degree should always be <= 1.

Extra-outputs have free degrees.

[tbittar]

The check is done only within the binding constraints, it is not forbidden within port-definitions.
But I agree with your remark, I should rationalize all checks using the degree computation. Moreover I am not sure this is correctly done for max, min, abs and so on. I will open a specific PR to adress your comment and robustify validation tests and also treating issue #226 which is not handled by this PR contrary to your comment. So if this is ok I propose to merge as is and to continue validation work in another PR

[aoustry]

Are you sure this file is necessary? Intuitively I would say that if we didn't need it for abs/floor/ceil/max, we don't need it for dual/rc.

I would say that we can just decide on the degree of the expression (variable bounds, constraint, binding-constraint or objective). This degree should always be <= 1.

[tbittar]

Comments handled in #229 , can we merge this one @aoustry ?