Port over a minimal virtual fitting approach from matlab (#147)

[ltetrel]

This adds the minimal implementation for the virtual fitting algorithm.

Close #147

[ltetrel]

Current state for this PR:

• @TongFuKitware was able to develop the functions necessary for this workflow to run, it was tested by loading output data from the old matlab code (test code + data sent to @ebrahimebrahim).
• we have the VirtualFit class skeleton in place, with its members callable as standalone functions.
• there are still discussions about which attributes for the class, and eventually some to put in other classes (ex. transducer stand-off geometry in Transducer class).
• extract_skin_surface is called in the __post_init__ function and stores the segmented skin (np.ndarray) in a new attribute, as well as the fitted interpolator. This to be able to call VirtualFit.run with different targets and not re-segmenting the volume.
• a skin mesh can be constructed as well with ConvexHull to replace the interpolator and segmented skin (list of points as np.ndarray), and benefit from normals (which can help during transducer pose search).
• we are still missing proper end-to-end workflow (interaction with Slicer scene)
• we are still missing tests with expected data