torch/distributed: move quantization ops to libtorch instead of libtorch_python (#77444)

Summary: This moves the distributed/c10d/quantization ops to libtorch instead of libtorch_python. Since they don't depend on Python this avoids the duplicate registration error from TORCH_LIBRARY when used in torch::deploy.

Test Plan:
Build

https://www.internalfb.com/intern/testinfra/testconsole/testrun/8162774414164820/

Differential Revision: D36384244

Pull Request resolved: https://github.com/pytorch/pytorch/pull/77444
Approved by: https://github.com/wanchaol, https://github.com/malfet
diff --git a/tools/build_variables.bzl b/tools/build_variables.bzl
index 0453852..4df8655 100644
--- a/tools/build_variables.bzl
+++ b/tools/build_variables.bzl
@@ -482,6 +482,7 @@
     "torch/csrc/distributed/c10d/exception.cpp",
     "torch/csrc/distributed/c10d/logger.cpp",
     "torch/csrc/distributed/c10d/logging.cpp",
+    "torch/csrc/distributed/c10d/quantization/quantization.cpp",
     "torch/csrc/distributed/c10d/reducer.cpp",
     "torch/csrc/distributed/c10d/sequence_num.cpp",
     "torch/csrc/distributed/c10d/socket.cpp",
@@ -944,7 +945,6 @@
 libtorch_python_distributed_core_sources = [
     "torch/csrc/distributed/c10d/init.cpp",
     "torch/csrc/distributed/c10d/python_comm_hook.cpp",
-    "torch/csrc/distributed/c10d/quantization/quantization.cpp",
 ]
 
 libtorch_python_distributed_sources = libtorch_python_distributed_core_sources + [