ROS integration for whole-body MPC

This PR adds ROS support for the whole-body MPC (my branch wjallet/devel). It eliminates a lot of dead code and makes necessary changes to some of the code architecture.

Changes:

• simplify code which creates Crocoddyl OCP (ocp_croccodyl.py)
  • use numpy masks where applicable
  • rename ambiguous "task" variable
  • do not use np.insert or np.append
• simplify the C++ Params class by:
  • removing a lot unused parameters (for centroidal MPC)
  • direct parsing of Yaml arrays to C++ Eigen vectors or matrices
  • remove or change type data members defined as std::vector only to be converted in Python to numpy
  • default constructor default-initializes all data members, does nothing else
  • static factory function create_from_file(filename=default_filename) replaces older ctor
  • add static factory create_from_str() which allows parsing from a std::string
• enable pickling for the Params class (relevant for multiprocessing on macOS apparently)
• fix the CMake config for macOS
• fix package.xml
• add CMake listfile formatting and linting
• enable conversion from C++ YAML::Node to Python dicts (for Python-only params)
• add utils for converting between ROS multiarrays and NumPy arrays
• change the TaskSpec class:
  • some parameters are loaded from the Yaml config file
  • renames file problem_data.py to task_spec.py
• rename CrocOCP to AlgtrOCP
  • separates AlgtrOCP into a subclass for instantiating ProxDDP, and one for proxddp.SolverFDDP
• update the README
• do not pass around the TaskSpec object
• python: remove qrw.Utils.init_robot() because it modified the Params object without telling the user
  • replace by more explicit and self-contained make_footstep()
  • no attributes added to Params object (params.T is replaced by params.N_gait in C++ and initialized there)

We have tested building with Linux, macOS, and macOS with Catkin from robostack.

Thanks to the work of Etienne Arlaud @Inria Paris.

Resolves #3