Duplicate dequant nodes for dq encoder (pytorch#4222)

Summary:
Pull Request resolved: pytorch#4222

{F1749810435}

Here, slice_9 and slice_10 are nops. When eliminated, the graph reduces to

{F1749810958}

dequantize_per_tensor_tensor_3 becomes input to two linear ops, causing failure in dq partitioner. This diff creates duplicates of dequant nodes so that each linear has unique inputs for dq partitioner.

Reviewed By: larryliu0820, pssrawat

Differential Revision: D59647685

fbshipit-source-id: 51ac80b11ce855a5eee2746c815d2f882beaa520

backends/transforms/TARGETS

@@ -120,6 +120,7 @@ runtime.python_library(
         "//executorch/backends/...",
         "//executorch/examples/...",
         "//executorch/extension/llm/...",
+        "@EXECUTORCH_CLIENTS",
     ],
     deps = [
         "//caffe2:torch",