Sounds great!

Here is how I see it working:

The merge button sends a merge/apply request event, which optionally specifies conflict resolution preferences.

A maintainer has set up a service that has an account with the git server which is authorised to push to push to main branch of the git repo. It watches for mergerequest event and on receipt does the following:

clones the repo and applies the patch against the tip of master. It either resolves the conflicts automatically and sends a merge/applied event (specifying the commit id) or sends a failed with unresolved conflicts event.

There could also be a DVM for running CI and reporting back but that doesn't need any authorisation from a maintainer.

What do you think?