[BUG] Not using GPU by default in RNNModel

[rocket-yg]

Describe the bug
I am trying to train a RNN model for a time series problem statement. Using Google Colab GPU backed instance, when I train, it doesn't use the GPU by itself as shown in training logs:

WARNING:darts.models.forecasting.torch_forecasting_model:DeprecationWarning: kwarg `verbose` is deprecated and will be removed in a future Darts version. Instead, control verbosity with PyTorch Lightning Trainer parameters `enable_progress_bar`, `progress_bar_refresh_rate` and `enable_model_summary` in the `pl_trainer_kwargs` dict at model creation.
GPU available: True, used: False
TPU available: False, using: 0 TPU cores

When i explicitly pass the 'cuda:0' string to the torch_device_str argument, I get a value error message:

ValueError: 'cuda' is not a valid DistributedType

Expected behavior
It should be able to train on the gpu instance.

[gsamaras]

darts version 0.17.0 was released yesterday.
The TorchForecastingModel are now built on top of PyTorch Lightning.

So try changing:

torch_device_str="cuda:0"

to:

pl_trainer_kwargs={
        "accelerator": "gpu",
        "gpus": [0]
}

Read more in https://pytorch-lightning.readthedocs.io/en/stable/common/trainer.html.