Add backwards compatibility for loading prediction head

[tholor]

In older versions of FARM the TextClassificationHead had the parameter "balanced_weights" which also got stored in the .bin. Loading these older models in the recent version of FARM gives:

  File "../torch/nn/modules/module.py", line 839, in load_state_dict
    self.__class__.__name__, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for TextClassificationHead:
	Unexpected key(s) in state_dict: "balanced_weights". 

This is due to torch.nn.module.load_state_dict(), which has a strict flag that handles if a model can be loaded even though there are different keys. Let's allow changing this flag from FARM's load methods to allow backward compatibility.