fix: torch_forecasting_model load_weights with float16 and float32 #4452

Job	Run time
check-examples (03-FFT-examples.ipynb)	3m 35s
check-examples (04-RNN-examples.ipynb)	7m 40s
lint	56s
check-examples (00-quickstart.ipynb)	6m 40s
tests (macos-latest, 3.9, all)	38m 26s
docs	8m 13s
check-examples (02-data-processing.ipynb)	6m 25s
tests (ubuntu-latest, 3.9, all)	46m 21s
check-examples (01-multi-time-series-and-covariates.ipynb)	7m 13s
	2h 5m 29s

Provide feedback