[cuDNN][cuDNN V8 frontend API] Clean up `time_sorted_plan` workaround for cuDNN v8 API (#100287)
`cudnn-frontend` (bumped in #99674) has added support for limiting the number of kernels to benchmark, so we can remove the workaround introduced in #91032.
CC @ngimel @ptrblck
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100287
Approved by: https://github.com/ngimel
