feat: Allow model version promotion without need for re-deployment

A really nice feature of mlflow is the ability to update registered model versions' tags to Production which then can be used to signal downstream dependencies to pull the newly promoted model version the next its cache invalidates (inference endpoints) or at the next run (batch ml job).

I believe it should be possible through metadata on the model's pin to be able to specify model version tags.

Then within the VetiverAPI or VetiverModel classes, check the metadata for the currently promoted model version and pull that one specifically instead of defaulting to the latest or requiring the dev to explicitly define it.

Would be super powerful to be able to update a deployed model just by updating the underlying pin's metadata.

Start by reading the metadata available on a model's pin and the VetiverAPI and VetiverModel entry points. Trace how the currently selected model version is resolved, then define how a promoted version tag should affect that choice without redeployment. Done means downstream inference or batch use selects the promoted version after metadata changes.