Most development shops are using the traditional client-server model, or self-host using Gitlab.

I personally would never allow a repo with two hashing algorithms to exist on my watch.

If you have ever had to use a tool like BFG to prune large objects from a repo you'll see it's not that bad, but it does require users to re-clone.

I would want to use the same process for SHA256 - that is let it be the default for new projects and then convert older projects based on need.

But there needs to be a BFG style conversion tool that spits out an object id map as output.

Here's more info on BFG: https://rtyley.github.io/bfg-repo-cleaner/