Argument to turn off strict model state dict loading checks

[MaxFBurg]

If one wants to load a pre-trained model and change some model config parameter, e.g. pre-trained with shifter, but then do experiments with shifter=False, this can be obtained by setting transfer=True in the model config. However, then model initialization fails, because trasfer usually is not an argument for our models. Could we refactor this to allow for model loading in a more flexible way? E.g. by adding another parameter to load_model that allows for turning off these checks, without having to modify the model_config?

nnfabrik/nnfabrik/builder.py

Lines 78 to 81 in 5b6e737

	match_names=model_config.get("transfer", False),
	ignore_unused=model_config.get("transfer", False),
	ignore_dim_mismatch=model_config.get("transfer", False),
	ignore_missing=model_config.get("transfer", False),

Might be related to #147