Update PyTorch THPVariable_Wrap to use const ref

[huydhn]

What does this PR do? Please describe:
This comes from a change from PyTorch pytorch/pytorch#138880 in which the signature of THPVariable_Wrap changes to

TORCH_PYTHON_API PyObject * THPVariable_Wrap(const at::TensorBase& var);

A build failure happens when importing the change to fbcode D65031075 for one project that is using fairseq2

Does your PR introduce any breaking changes? If yes, please list them:
This makes fairseq2 compatible with latest PyTorch, so I don't think there is any breaking change here.

Check list:

• Was the content of this PR discussed and approved via a GitHub issue? (no need for typos or documentation improvements)
• Did you read the contributor guideline?
• Did you make sure that your PR does only one thing instead of bundling different changes together?
• Did you make sure to update the documentation with your changes? (if necessary)
• Did you write any new necessary tests?
• Did you verify new and existing tests pass locally with your changes?
• Did you update the CHANGELOG? (no need for typos, documentation, or minor internal changes)

[cyyever]

@huydhn Should this project use latest release of torch instead of the developing branch?

[huydhn]

I think we have a chick-and-egg issue here. So here is the current interlocking setup:

• Some Meta-internal code uses pytorch (latest) and fairseq2 (imported from this repo). They fails because these twos are not compatible anymore w.r.t THPVariable_Wrap
• fairseq2 is open source and it tests against stable PyTorch >= 2.2.0 https://github.com/facebookresearch/fairseq2/blob/main/.github/workflows/ci_build_wheels.yaml#L33-L52. This is not latest PyTorch.

I think I need to try writing an internal patch instead of waiting for this PR.

[huydhn]

I will close this for now as the internal patch works (but brittle). I'll leave it to fairseq2 maintainers to decide when and how they want to make fairseq2 compatible with PT when it comes to this public function