bug: fix accuracy metric and add `test_with_y=True`

[EllingtonKirby]

Hi there! Thank you so much for your work on this project. I was using your script to finetune ELECTRA on a smaller dataset and noticed that the learner was not reporting accuracy or validation loss. I took some time to look into it and figured out the issue on my end.

The first issue was that without the test_with_y=True parameter, I did not see the labels inside the examples of the validation data loader. So I have added this parameter.

Second, the implementation of accuracy in the FastAI API used in the project seems to be broken, it was throwing an error like AttributeError: 'bool' object has no attribute 'float', so I have added code to cast the two tensors. I'm not sure this is the best way to do this but it worked for me.

Thanks again for your work