fix aten.reflection_pad2d (#9289)

tested the torch doc example

extra/torch_backend/backend.py

@@ -311,8 +311,7 @@ tiny_backend = {**{k:wrap_out(v) for k,v in tiny_backend_out.items()}, **{
   "aten.logical_not": Tensor.logical_not,
   "aten.masked_fill_.Scalar": lambda self,mask,value: self.assign(mask.where(self, value)),
   "aten.multinomial": Tensor.multinomial,
-  # TODO: this is wrong
-  "aten.reflection_pad2d": Tensor.pad,
+  "aten.reflection_pad2d": functools.partial(Tensor.pad, mode="reflect"),
 }}
 
 def wrap_fxn(k,f):