refactor "algorithm" debug ?

[bpinsard]

The use of algorithm view is necessary to adjust the pupil intensity threshold.
I think there would be added value to return the algorithm debug info as objects rather than embedding it in a color image.
The advantages would be :

• avoid having to convert to rgb when the native frame is not rgb (grayscale, yuv, ...), for instance just np.tile to convert from our gray input to rgb is making a 250 fps to drop dramatically, causing frame loss/lags/overflow. ( Can't use numpy tricks as the array needs to be c-contiguous.)
• display debug in pupil by drawing over the captured image, prettier when windows scaled, and likely faster than rasterizing on CPU.
• allowing external program to analyze the debug info.

WDYT?