Improve documentation for output parameters expected to be set at 0

[EtienneAr]

Hello,
I realized recently that some methods expect the ouptut parameter to be initialized to 0. Such as:

pinocchio/include/pinocchio/algorithm/jacobian.hpp

Line 173 in 187afaf

void computeJointJacobian(

I don't think it is very obvious for the user since it's only said once in the middle of a long docstring:

pinocchio/include/pinocchio/algorithm/rnea-second-order-derivatives.hpp

Lines 13 to 70 in 187afaf

	///
	/// \brief Computes the Second-Order partial derivatives of the Recursive Newton
	/// Euler Algorithm w.r.t the joint configuration, the joint velocity and the
	/// joint acceleration.
	///
	/// \tparam JointCollection Collection of Joint types.
	/// \tparam ConfigVectorType Type of the joint configuration vector.
	/// \tparam TangentVectorType1 Type of the joint velocity vector.
	/// \tparam TangentVectorType2 Type of the joint acceleration vector.
	/// \tparam Tensor1 Type of the 3D-Tensor containing the SO partial
	/// derivative with respect to the joint configuration vector. The elements of
	/// Torque vector are along the 1st dim, and joint config along 2nd,3rd
	/// dimensions.
	/// \tparam Tensor2 Type of the 3D-Tensor containing the
	/// Second-Order partial derivative with respect to the joint velocity vector.
	/// The elements of Torque vector are along the 1st dim, and the velocity
	/// along 2nd,3rd dimensions.
	/// \tparam Tensor3 Type of the 3D-Tensor
	/// containing the cross Second-Order partial derivative with respect to the
	/// joint configuration and velocty vector. The elements of Torque vector are
	/// along the 1st dim, and the config. vector along 2nd dimension, and velocity
	/// along the third dimension.
	///\tparam Tensor4 Type of the 3D-Tensor containing the cross Second-Order
	/// partial derivative with respect to the joint configuration and acceleration
	/// vector. This is also the First-order partial derivative of Mass-Matrix (M)
	/// with respect to configuration vector. The elements of Torque vector are
	/// along the 1st dim, and the acceleration vector along 2nd dimension, while
	/// configuration along the third dimension.
	///
	/// \param[in] model The model structure of the rigid body system.
	/// \param[in] data The data structure of the rigid body system.
	/// \param[in] q The joint configuration vector (dim model.nq).
	/// \param[in] v The joint velocity vector (dim model.nv).
	/// \param[in] a The joint acceleration vector (dim model.nv).
	/// \param[out] d2tau_dqdq Second-Order Partial derivative of the generalized
	/// torque vector with respect to the joint configuration.
	/// \param[out] d2tau_dvdv Second-Order Partial derivative of the generalized
	/// torque vector with respect to the joint velocity
	/// \param[out] dtau_dqdv Cross Second-Order Partial derivative of the
	/// generalized torque vector with respect to the joint configuration and
	/// velocity.
	/// \param[out] dtau_dadq Cross Second-Order Partial derivative of
	/// the generalized torque vector with respect to the joint configuration and
	/// accleration.
	/// \remarks d2tau_dqdq,
	/// d2tau_dvdv, dtau_dqdv and dtau_dadq must be first initialized with zeros
	/// (d2tau_dqdq.setZero(), etc). The storage order of the 3D-tensor derivatives is
	/// important. For d2tau_dqdq, the elements of generalized torque varies along
	/// the rows, while elements of q vary along the columns and pages of the
	/// tensor. For dtau_dqdv, the elements of generalized torque varies along the
	/// rows, while elements of v vary along the columns and elements of q along the
	/// pages of the tensor. Hence, dtau_dqdv is essentially d (d tau/dq)/dv, with
	/// outer-most derivative representing the third dimension (pages) of the
	/// tensor. The tensor dtau_dadq reduces down to dM/dq, and hence the elements
	/// of q vary along the pages of the tensor. In other words, this tensor
	/// derivative is d(d tau/da)/dq. All other remaining combinations of
	/// second-order derivatives of generalized torque are zero. \sa
	///

Moreover, the way to say it is not always consistent between every function that requires it. (For instance, here is slightly different: )

pinocchio/include/pinocchio/algorithm/jacobian.hpp

Lines 152 to 153 in 187afaf

	/// \param[out] J A reference on the Jacobian matrix where the results will be stored in (dim 6 x
	/// model.nv). You must fill J with zero elements, e.g. J.setZero().

What could be done to smooth the user experience ?

I was thinking, as a first step, to add asserts in those methods to check in debug that the parameter in indeed set to 0.
Do you want me to open a PR ?