[quantization] Make FP16 RNN use new prepack op (#34339)
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/34339
Test Plan: Imported from OSS
Differential Revision: D20297194
Pulled By: jamesr66a
fbshipit-source-id: 8bf6d0f2cb047e90bbdd184aaad337b143040d10
