Review what should be returned from ops that return view tensors

There are some Symbols that return a new tensor while the corresponding PyTorch ops return a view.
The set of such ops is defined here -- https://github.com/Lightning-AI/lightning-thunder/blob/62d9535bfeab621b571c3cc8aed974b64381aad0/thunder/torch/__init__.py#L6634-L6660.

For those ops, it might be beneficial to update TensorProxy so that we can tell the outputs of such ops share underlying storage with some other TensorProxy's, then it'd be reasonable to expect improved support for in-place ops on those view tensors.

Status Quo: Returning a new tensor from those ops isn't 100% correct but in-place functionalization and alias updates for in-place both special-case those ops -- https://github.com/search?q=repo%3ALightning-AI%2Flightning-thunder+syms_returning_views&type=code

This issue derives from Mike's review on #2213:

Won't this be incorrect with inplace operations, because this will create a copy, but if the result of the view is modified inplace then the original tensor should be changed?

Originally posted by @mruberry in https://github.com/Lightning-AI/lightning-thunder/pull/2213#discussion_r2138237788

Start with the syms_returning_views definition in thunder/torch/init.py around lines 6634-6660, then inspect the in-place functionalization and alias-update code found by the repository search. Trace how TensorProxy represents outputs from these operations and determine what changes are needed for view outputs to share storage correctly. Done means the affected operations have correct view semantics and improved in-place support, with the relevant behavior covered by tests if existing tests are identified.