Raise Error when user opts for DDP on a system without CUDA

Currently, if the user sets `use_data_parallel=True` in `ConfigurableTrainer` on a machine without a GPU, the code simply crashes (as is expected). But the error message is misleading. Handle this with an assertion with a clear message.

Start by locating ConfigurableTrainer and the handling of use_data_parallel=True. Reproduce the current behavior on a machine without CUDA, then verify that the assertion reports clearly why DDP cannot be used; done means the misleading crash is replaced by that assertion.

Written by the indexing model from the issue text.