Allow Object Detector ROI to guide ISP exposure

[Luxonis-Brandon]

Start with the why:

When doing object detection, often the object (once found) is of most interest. So having the ROI of the object detector (e.g. the highest-confidence object class of interest) would allow the object detector to increase the probability that it continues to find the object as lighting/etc. changes.

Move to the how:

Implement an optional feedback hook in the ISP processing such that the ROI that is parsed in ISP is purely from the object detector ROI.

Move to the what:

Object-Detector-based exposure for better object detection in variable lighting conditions.

[Luxonis-Brandon]

This is now implemented and has an example of how to do so, here: https://github.com/luxonis/depthai-python/pull/113