Improve RANSAC Ground Plane Extraction

[niwhsa9]

Once the model with the highest inlier count is extracted, it may be beneficial to perform linear least-squares amongst those inlier points to refine the plane. Then, re-select inliers from the entire dataset. This can actually be performed iteratively to continue ground plane refinement.

Code for selecting inliers given a model already exists in the ransacKernel and is quite fast. The major add would be solving a system Ax=b, which SHOULD be able to be solved very quickly with QR factorization using the cusolver library: https://docs.nvidia.com/cuda/cusolver/index.html. The matrix is dense so be sure to use the dense solver.

[niwhsa9]

actually, lets start with using two different thresholds for removal and determining inliers, this may be able to solve the issue