don't use prims.unsqueeze in group_norm (#88927)
inductor doesn't have prims.squeeze lowering, so this breaks it. Longer term, `squeeze` with multiple dimensions is not a prim, nvfuser implements it with a loop, inductor uses `_squeeze_multiple` helper which turns it into a loop. Prim should accept only a single dimension.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/88927
Approved by: https://github.com/eellison
