Allow integrating general-purpose FMUs into CasADi expressions with derivative information exploited:

• Loading FMU DLLs
• Avoid reinitializing FMU when not necessary (i.e. when called inside an ODE integrator)
• Exploit Jacobian sparsity information
• FMU specific finite differencing support, exploiting information about variable bounds and nominal values, as well as the ability to reevaluate FMUs cheaply if only a small subset of the input changes
• Validation of AD using FD support
• Automatic step size selection for FD
• Support different FD schemes
• Graph coloring support for Jacobian calculation
• Smaller buffers in Fmu, including only relevant subset of variables in FMU
• Make FmuFunction independent of DaeBuilder after creation
• Parallel Jacobian/adjoint calculation - OpenMP
• Parallel Jacobian/adjoint calculation - std::thread
• Support for forward directional derivatives cf. Forward sensitivities in FMU interface #3054