* added tff.py for TFFModel

* savepoint switching to master

* TFT with MixedCovariates

* savepoint software update

* checked tft_submodels

* original model performs well

* all submodels checked and ready

* new tft submodels work as original TFT

* github test

* adapted forward for fitting with darts

* predict returns output but issue with target scaling

* fixed bug in _produce_train_output(), now predictions look much better

* train and prediction work

* TFT now supports variable prediction lengths

* added multivariate TFT forecast support

* added model docstrings

* general clean up of unnecessary variables and code

* removed files used for building the model

* added TFTModel to README.md

* added probabilistic forecasting support for TFT with likelihood models

* minor docstring fixes

* fixed error from testing with MixedCovariatesInferenceDataset for regression models

* removed unused changes in darts/utils/data/inference_dataset.py

* Revert "removed unused changes in darts/utils/data/inference_dataset.py"

This reverts commit a8521cc545007cafdeab184a281595475cdf020f.

* remove unused changes in darts/utils/data/inference_dataset.py

* added TFT unit tests

* cleaned up unused features in TFT unit test

* fixed *args to predict() method for model.historical_forecastings()

* solve an issue with Cuda

* made compatible with pandas==1.3.0 and statsmodels==0.13.0

* added package pytorch-forecasting>=0.9.1

* added submodels from pytorch-forecasting

* removed pytorch-forecasting dependency and applied part of PR review suggestions

* Feat/quantile loss as likelihood (#526)

* applied changes from PR review

Co-authored-by: Julien Herzen <julien.herzen@unit8.co>
Co-authored-by: Julien Herzen <j.herzen@gmail.com>