Casadi

Changes

Major
- implemented support for discrete variables (see argument `discrete` in `csnlp.Nlp.variable`)
- implemented new wrapper `csnlp.wrappers.PwaMpc` for creating model predictive control (MPC) controllers for piecewise affine (PWA) systems. The class takes care of automatically converting the PWA dynamics to a big-M mixed-logic relaxation
- renamed `csnlp.wrappers.Mpc.set_dynamics` to the more informative name `csnlp.wrappers.Mpc.set_nonlinear_dynamics`
- implemented `csnlp.wrappers.Mpc.set_linear_dynamics`
- all dynamics constraints are now generated via `casadi.Function.map` (in multiple shooting) and `casadi.Function.mapaccum` (in single shooting) for faster construction time

Minor
- implemented new examples for mixed-integer optimization problems and optimal control problems
- better docstrings and error messages
- dynamics functions are no longer saved, after setting them, in the `csnlp.wrappers.Mpc` class and subclasses

Changes

Major
* Implemented a lazy version of `csnlp.Solution` that will postpone computations of various quantities pertaining to a primal-dual optimal point until necessary
* Removed the need to specify `n_in` and `n_out` in `csnlp.wrappers.Mpc.set_dynamics`
* Refined the dependency on NumPy


Minor
* In `csnlp.wrappers.NlpSensitivity`, the computation of Jacobians and Hessians was streamlined: now only those sensitivities that are needed by the user are computed
* Bug fixed in `csnlp.Nlp.init_solver` where the solver options were not properly saved for later re-initalizations
* Improvements and fixes to documentation

Changes

- Now documentation is available online [here](https://casadi-nlp.readthedocs.io/en/stable)
- Updated dependency to `numpy < 2.0.0`, as the newest release of `numpy` breaks `casadi`
- Updated dependency to `joblib >= 1.4.0`
- Updated `multistart.ParallelMultistartNlp.__init__` to accept a generic `parallel_kwargs` dict to instantiate the `joblib.Parallel` backend

Changes

Major
- added support for NumPy's .npz saving format to the I/O utils in `csnlp.util.io`
- implemented new multistart NLP paradigm, the `MappedMultistartNlp` class, which leverages the `casadi.Function.map` API

Minor
- improvements to `ParallelMultistartNlp`: now the solutions are ranked first by feasibility, then by success, and only at the end by optimality
- improvements to `RandomStartPoints`: now supports two new arguments, `biases` and `scales`
- improved internal logics and type hints

Changes

Major
- fixed error on bounds on initial and terminal states in `ScenarioBasedMpc`

Minor
- improvements to internal files
- switched to pre-hooks
- fixes to tests
- better type hinting for seed

Changes

Major
- removed support for Python 3.8
- implemented the `ScenarioBasedMpc` wrapper to create MPC controllers based on the Scenario Approach
- added extra args in `Mpc` and `ScenarioBasedMpc` to allow to include or exclude lower and upper bounds on initial and terminal state upon creation
- removed `quad_form` in favour of `casadi.bilin` (which is built in in the library)
- added support for all solvers (previously, only IPOPT, sqpmethod, and a small other few could be instantiated)

Minor
- updates to config files
- updated plotting utilities
- moved from `full` to `toarray`, which appears to be faster
- moved to relative imports