Use point wise real time

When undistortion is enabled, we use the point-wise time field as a frame advancement to try to estimate the pose at the beginning and end of the scan, in order to undistort the full scan.

To do that, we normalized the point-wise time in range [0-1], and estimated the poses a time 0 and 1. This is clearly not needed : we can directly use the time value. By using it directly, it moves a step forward to use multiple sensors clouds by easing their time management.

This MR also clarifies the timestamp at which the output Tworld pose is computed. Tworld is now the pose at the timestamp held in the input scan header. It is not anymore the pose "at the end of the scan" as claimed before (this was not always true). If undistortion is enabled, we still estimate the poses at the beginning and end of the frame, then we interpolate Tworld using these 2 poses.

This should not change the behavior when undistortion is disabled. I noticed a much better behavior than before with undistortion mode OPTIMIZED. In APPROXIMATED mode, the behavior looks similar.

This MR also clarifies and simplifies a lot the undistortion process and robustness.

@julia.sanchez @nick.laurenson @lea.vauchier @melanie.carriere PTAL