Update how to build PyTorch with CUDA Windows instructions

[malfet]

PyTorch currently could not be build using recommended 14.11.25503 minimal toolchain, see:
https://github.com/pytorch/pytorch/blame/b4b0fa637178baf9147416b550c7db70de6a5fa3/README.md#L258

But if one tries to following this instructions using PyTorch-1.7 or newer it will fail with as shown in:
#46208 (comment)

cc @malfet @seemethere @walterddr @jlin27 @mruberry @peterjc123 @maxluk @nbcsm @guyang3532 @gunandrose4u @smartcat2010 @mszhanyi

[mszhanyi]

From the issue, the user built with cuda 10.0, but cuda 10.0 looks not compatible with vs 2019.
So we need tell users that they have to clone python 1.6 or lower if they want to use cuda 10.0 or lower, am I right?

[peterjc123]

@mszhanyi I think we should just say that we support cuda >= 10.1 and vs >= 2019.

[peterjc123]

So we need tell users that they have to clone python 1.6 or lower if they want to use cuda 10.0 or lower, am I right?

Maybe we could create a new table for this. Something like

CUDA VERSION	CUDA ARCH	LATEST PYTORCH VERSION SUPPORTED
10.0	3.5+	1.6.0
9.2	3.5+	1.x.0

[skyline75489]

#42722 (comment) is another compilation error caused by VS2017 with 14.16.27023 toolchain.

I guess it's time to drop the support of VS2017.