OCR: Investigate the huge discrepancy between reward proportions

So it turns out that when Nihar uses the snapshots to calculate reward proportions on mainnet-beta epoch, the proportions are wayy off between his R/python code vs what I do in the rust contributor-rewards code.

This is unacceptable, I have _some_ idea why this might be happening but I need to dig further. This just moved to be my Priority#1.