ci: Uncap torch, allowing it to bump to 2.7.1 in constraints

[booxter]

This also brings newer CUDA into CI.

Signed-off-by: Ihar Hrachyshka [email protected]

[booxter]

first merging #612 before considering this PR, due to overlap.

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. @booxter please rebase it. https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[booxter]

To clarify the rationale here: the main product that consumes this library already switched to newer torch and (even newer) CUDA, we are past the point of considering holding testing in gate for newer versions. I know there were arguments made that a lot of leg-work should be done when bumping versions (for example, suggestions to run extensive result comparisons or looking at specific code diff between versions) but I don't think the onus of doing it is on folks taking care of CI. If training team is willing to invest in this, it's honorable of course (though I think looking at actual code diff is security theater and unnecessary slowdown; additional monitoring / analysis may be a better approach).

[github-actions]

E2E (NVIDIA L40S x4) (python 3.11) workflow launched on this PR: View run

[github-actions]

e2e workflow succeeded on this PR: View run, congrats!