Centroid magnitude grounded in reality

[markasoftware]

(This is probably for post-smallsat submission)

Realistically, we should do something like: logarithm of sum of (pixel brightness - threshold), since this somewhat corresponds to how magnitude is actually computed.

Until this is implemented, centroid magnitude should refer to the number of pixels above threshold. And in the long run, star-id algos should only assume that magnitude is somehow correlated to brightness.