Avoid copies in matmul (no ghstack) (#97355)
Resubmit of https://github.com/pytorch/pytorch/pull/76828 without using ghstack so that @ngimel can import it and help me debug the issue why it was reverted.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/97355
Approved by: https://github.com/ngimel, https://github.com/malfet
