Use correct full-op in replace_scalar_with_tensor pass (pytorch#16637)

Use torch.ops.aten.full.default for torch ops and
exir_ops.edge.aten.full.default for exir ops.

Signed-off-by: Adrian Lundell <adrian.lundell@arm.com>

Author: AdrianLundell

backends/transforms/replace_scalar_with_tensor.py

@@ -11,6 +11,7 @@
 from executorch.exir.dialects._ops import ops as exir_ops
 from executorch.exir.dialects.edge._ops import EdgeOpOverload
 from executorch.exir.pass_base import ExportPass
+from torch._ops import OpOverload
 
 
 class ReplaceScalarWithTensorArgPass(ExportPass):
@@ -46,6 +47,11 @@ def __init__(
         super().__init__()
 
     def get_replacement(self, op, args, kwargs, meta):
+        if isinstance(op, OpOverload):
+            full_op = torch.ops.aten.full.default
+        else:
+            full_op = exir_ops.edge.aten.full.default
+
         return super().call_operator(
             # Replace with .Tensor variant.
             op=self.scalar_to_tensor_ops[op],
@@ -54,10 +60,10 @@ def get_replacement(self, op, args, kwargs, meta):
                 args[0],
                 # Scalar arg - replace with aten.full tensor.
                 super().call_operator(
-                    exir_ops.edge.aten.full.default,
+                    full_op,
                     args=(
                         (1,),
-                        args[1],
+                        float(args[1]),
                     ),
                     kwargs={
                         "dtype": args[0].to_tensor().dtype,