Is it possible to have an application where depending on the change in frequency with time, the sampling also keeps changing?

yes, many models for video understanding have different temporal sampling rate.

Do people actually try to find the highest frequency in the image, or is it more common to just try and sample as fast as we can?