It seems this was considered: https://git-scm.com/docs/hash-function-transition/#_using_ha...

The biggest one of those downsides I think is that the SHA-1 history could be changed out from under you if a new attack was published.