Adding online learning to lds model

[ncguilbeault]

This PR adds online learning capabilities to the existing linear dynamical systems model for kinematics.

Changes to main.py

Several modifications were made to the main.py code. The KalmanFilterKinematics class now contains several new attributes and functions. Critically, the optimize function calls the lds.learning.scipy_optimize_SS_tracking_diagV0 function from the lds package to perform optimisation of the selected parameters. The run_optimization function batches the observations and calls the optimize function. The run_optimization_async function performs the same way, except it creates a separate thread to run optimization and sets a callback function called on_completion to report back when optimization has finished. There are additional functions to check when optimization is finished and to reset the optimization loop.

Include workflows added

Two additional IncludeWorkflows were added. The RunOptimizationAsync workflow allows the model to perform asynchronous optimization of selected model parameters given mini batches of data. The workflow will produce a single output once the optimization procedure has started. Therefore, to check when the optimization has finished, the CheckOptimizationFinished workflow tests on every input whether the model is currently performing optimization and emits a value when the optimization procedure has completed.