Specifying +cuXXX and +cpu

[Yoinky3000]

Description

For some reason uv cannot resolve torch dependencies correctly unless +cuXXX/+cpu is specified in the torch command env

when I'm installing for cuda without +cuXXX:

after I added it:

the lint gha in this repo(with uv enabled):

Notes

UV docs explained about this kind of situation here:
https://docs.astral.sh/uv/pip/compatibility/#local-version-identifiers

I only edited the torch command for CUDA and CPU only, as I'm not sure about AMD and Intel gpu

Environment and Testing

List the environment you have developed / tested this on

[vladmandic]

its not just uv, its a major issue overall.

e.g. some packages have torchvision as dependency, but python does not resolve installed torchvision+cu same as torchvision, so it looks like there is no torchvision already installed so it pulls in latest which pulls in latest torch which invalidates torch_command to start with. torch versioning is very well known not to be inline how python versioning is done and it breaks many things.

but ok, i'm wiling to give this a try, i doubt it can make current mess any worse.