Option to extract a number of video frames as a function of video duration

[lfcnassif]

As discussed in other issues #1081 #1102, this could improve search results of a number of video tasks (nudity and faces detection, future similarity search, OCR and photoDNA and others).

extractFramesMillisecondInterval = 1000 would be a good default value for pedo and forensic profiles and disabled for others?

Or should we make some kind of attenuation targeting huge videos, maybe using a logarithmic or similar function applied to video duration?