remove unused tensors from VK model's graph (#4427)

Summary:
Pull Request resolved: #4427

We implemented [operators fusion](#3769) (`conv+bn`) which fused `conv` and `bn`'s weights and biases, but the old parameters are not deleted. Hence we saw that VK model's size is nearly twice large as CPU's.

As regards mobilenet_v2, before this diff CPU vs VK: 14M vs 22M. After this diff, both of them have 14M.

Reviewed By: SS-JIA

Differential Revision: D60257047

backends/vulkan/serialization/vulkan_graph_builder.py

@@ -262,6 +262,9 @@ def get_or_create_value_for(self, arg: _Argument):
             raise RuntimeError(f"Cannot create value for arg of type {type(arg)}")
 
     def process_placeholder_node(self, node: Node) -> None:
+        # ignores any tensors that don't get used in any ops
+        if len(node.users) == 0:
+            return
         ids = self.create_node_value(node)
         if not self.is_param_node(node):
             if isinstance(ids, int):