Accuracy doesn't work for simple binary classifier

[Richardjmorton]

Hello,

This is quite a simple issue but the function nnx.metric.Accuracy only works for a network that returns more than one output, i.e.
super().update(values=(logits.argmax(axis=-1) == labels)). Not sure if this is by design.

I was playing around with a simple binary classifier than only returned a single logit per observation. The Accuracy function gives nonsense for this case.

Anyway, might be worth noting in the doc string what the function expects.