Updated internal code for orgqr function (#56247)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/56247
Moved `apply_orgqr` to `BatchLinearAlgebraKernel.cpp`.
Removed `infos` tensor parameter. We don't need to expose
lapack/cusolver error codes because they do not contain any useful
information about the input. Its value is checked only in debug mode now
removing the device syncronization from the cuSOLVER path of
`torch.linalg.householder_product` or `torch.orgqr`.
Test Plan: Imported from OSS
Reviewed By: ngimel
Differential Revision: D27844339
Pulled By: mruberry
fbshipit-source-id: 47aa20dfe2c116951b968362ad55e837caece042
