pytorch

c3861649 - Conservatively move all suitable prim ops from full-jit to mobile, and make them selective. (#58353)

Conservatively move all suitable prim ops from full-jit to mobile, and make them selective. (#58353) Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/58353 There are long tail operators in register_prim_ops_fulljit.cpp, that may be used in mobile build. In this PR 1. All of the ops that are likely to be used in mobile are moved to register_prim_ops.cpp. 2. Note that this move is conservative. If an op is likely to have fulljit dependency, or cannot be selective, it will be kept. Later if there's need to be used in mobile (rare), it will be adapted and moved case by case. 3. All the moved ops are marked selective. The registration function is changed from `Operator()` to `OperatorGenerator()`. Size regression is not expected. Test Plan: * Internal size tests * CI Reviewed By: dhruvbird Differential Revision: D28463158 Pulled By: iseeyuan fbshipit-source-id: 34536b8a569f1274329ccf1dac809fe9b891b4ff