Feature Request: Add "--unet-support-both" option to torch2coreml

[jrittvo]

It seems that an app implementation can choose between using Unet.mlmodelc or ControlledUnet.mlmodelc at model loading, or perhaps at inference run time. This means that a single model folder can contain both Unet versions, rather than needing what would appear to be a largely duplicative second model folder. As such, it would be convenient if torch2coreml had an argument to convert both Unet versions in a single run. Presently, one run is required to obtain eachUnet type, although the second run can be set to only provide the ControlledUnet.mlmodelc. If the user does not anticipate needing both Unet types, they would not use this added argument.