https://www.pnas.org/content/117/3/1261
Basically, the question everybody is interested in is P(shot|attributes). The database of shootings they used only lets them estimate P(attributes|shot). But because they don't have data on the rate of police/attribute encounters, they can't support any claims about P(shot|attributes).
They want to retract their study because they feel its information is being "misused," which is extremely disturbing in the context of honest academic research. I won't go into this deeply, but valid information on such a hot button subject is exactly what we need right now, and withdrawing valid information only leads to more vacuum, which will be filled by whatever the dominant political power declares.
As to your second statement, statisticians regularly use one quantity as a proxy for another. Here, they used the violent crime (by race) rates as a proxy for police encounters. This is in contrast to using gross population race proportions as a proxy for police encounters. Other studies--which have not been retracted--have used gross population race proportions as a proxy for police encounters. Will those be retracted? Will future studies using similar proxies be retracted? Only if they go against a powerful narrative, I wager.