Update `validate_panel` normalization method to use cohort-wide constants per channel

**Describe the bug**

Currently, the `validate_panel` function normalizes each channel separately based on its own 99.9% value. This will accentuate noise for very dim images, which leads to undesirable images.

**Expected behavior**

The normalization value needs to reflect the cohort as a whole, and not just on an individual level. To address, figure out what the best constants to use for each channel are, and use that to normalize all images across a cohort or sub-cohort.