Normalize and pil link fix

[kohr-h]

Update for a broken link to the PIL docs, and a more efficient version of the normalize transform.

[soumith]

thank you!

[vfdev-5]

@kohr-h looks good, as you are on F.normalize, maybe it would be nice to add a notice in the documentation to cover #517 ?
cc @soumith

[kohr-h]

@vfdev-5 What would be the intention here, efficiency or keeping input intact? Looking at other transforms, they never explicitly copy, but probably most of them can't work in-place anyway.

I can add a notice in the docs, sure.

[vfdev-5]

@kohr-h I would say that efficiency is better and among all other transforms F.normalize is the only transform working on torch.Tensor and returning torch.Tensor, so just add a notice that it works in-place can be enough IMO