One of the nice things about using Deeplabcut is that because of the way it processes images, its performance can work across many different cameras as long as you label some example data from each camera. For example, some of my sessions were recorded with color, some in black and white, and some with nice cameras vs. some with a relatively cheap webcam, and one single DLC model works well across all of them.
So I think your most important considerations as far as what camera to use should be sharp enough images for what your specific needs are - e.g. maybe you need to track fingers individually, versus just where hands are generally. The tradeoff here is that high-resolution videos take longer to process, and you need to have space to store long videos (adds up very quickly in my experience!)
(P.S. Less of a camera issue and more of a session setup consideration, but one thing to keep in mind is that giving labeled training data from many different angles will help to keep the labeling stable across future datasets!)