Periodic resample operation gradients and optimization (#16520)

* Implement gradient of periodic resample operation.

* Set fully defined output shape for periodic_resample when possible.

* Speed up periodic_resample operation.

Use incremental updates in index computation where possible.

* Allow periodic_resample run on multiple CPU kernels.

* Small refactoring.

* Add test for periodic_resample shape inference.

* Fix issues after review.

* Add shape inference C++ test.

* Code style fix