Tracking features in valid pixels only after rectification

[jingnanshi]

Description: Currently, Kimera-VIO detects features in all pixels of the image after rectification. However, for stereo cameras with non-standard configurations, the valid pixels after rectification will only occupy a limited region of the entire frame.

Potential Solution: Obtain the valid ROI output from the stereoRectify function of OpenCV: https://docs.opencv.org/2.4/modules/calib3d/doc/camera_calibration_and_3d_reconstruction.html#stereorectify. However, for fisheye cameras, we might need to do something different since the stereoRectify function does not return ROI: https://docs.opencv.org/3.4/db/d58/group__calib3d__fisheye.html#gac1af58774006689056b0f2ef1db55ecc