Add exact spectral norm feature

[lockwo]

Addresses part of https://github.com/patrick-kidger/equinox/issues/810.

[lockwo]

Open to discussion on best interface. The current caution around bias is ok but could be a footgun (although the original paper doing the reshaping is already kind of a footgun imo lol), but maybe I can apply the linear transpose around the whole layer pytree and update the bias too? I'm not sure that would 100% generalize to all layers though.