Currently, in case the targeted collection is timeseries we translate the command request to convert the index spec from logical to raw. Since we do not hold any collection lock during and afte the translation, it could happen that the collection gets droppe and re-created with a different collection type (timeseries -> normal or normal -> timeseries).
To spot this situation we currently set the UUID of the collection while translating the request and later on when we acquire the collection we need to re-check of the UUID changed.

We could avoid this by pushing down the translation after the final collection acquisition. So instead of translating the request at top level command entry point function we would simply convert the index spec after acquiring the collection.