This PR adds a pybind11-based pycddp package with bindings for solver options, dynamics, objectives, constraints, and solver execution, along with Python integration tests and packaging metadata.

It also makes Matplot-dependent includes and test links optional, adds CMake package installation metadata, and prepares the core library for shared-library consumers such as the Python module.

This keeps matplot and CDDP_HAS_MATPLOT available to in-tree builds while removing them from the installed cddp target interface, which avoids the cddpTargets export error.

It also cleans the Docker build step to use explicit CMake and CTest commands and removes the unused Python executable flag.

This updates the installed CMake and Python package layouts so downstream consumers can find cddp headers and import pycddp after install.

It also replaces the unsafe raw-pointer Python ownership handoff with Python-backed solver adapters, acquires the GIL for Python callbacks, and removes the legacy Matplot-driven build surface from the default configuration.

This raises Python-facing errors for invalid solver names and initial trajectories while preserving the existing C++ API behavior.

It also releases the GIL around solve(), rethrows threaded callback failures, and adds regression tests for nonlinear objectives and solution fields.

This restores parallel forward-pass behavior so a successful alpha still wins when another trial throws.

It also adds a Python Constraint trampoline plus regression coverage so custom path constraints can be defined and dispatched through the solver.

Improve Python import/autodiff error reporting and clarify solver ownership docs.
Handle parallel derivative precompute exceptions after draining worker futures,
and delete unreachable Matplot test code instead of preserving dead guards.

Force INSTALL_GTEST off before fetching googletest so enabling CDDP_CPP_BUILD_TESTS does not install the bundled dependency alongside cddp.

Release Python-owned wrapper references under the GIL so bound solver components are cleaned up safely. Also log partial forward-pass thread failures in verbose mode to make solver issues easier to diagnose.

Update the regression assertion to match the current actionable\nbinding message raised when set_initial_trajectory is called\nbefore a dynamical system is configured.

Pin pybind11 to <3 to avoid incompatibility with older bindings.

Install the Python extension and package files into the pycddp target directory directly.

Wire the augmented Lagrangian DDP solver into the core library, expose its options in the Python bindings, and register it as a built-in solver type.

Add focused ALDDP coverage for constrained, infeasible-start, and square-root backward-pass scenarios.

Reuse solver state across repeated constrained solves so warm starts
can keep their gains, and refresh ALDDP slack controls when a new
trajectory seed is supplied.

Simplify LogDDP barrier evaluation for single-shooting mode and add
regression tests for reseeded trajectories and invalid rollout options.