Update on "Traceable wrapper subclass support for deferred runtime as…

…serts"


The padded dense -> jagged conversion op has the signature:
```
_fbgemm_dense_to_jagged_forward(Tensor dense, Tensor[] offsets, SymInt? total_L=None) -> Tensor
```

when `total_L` is not specified, the meta registration has a data-dependent output shape (based on `offsets[0][-1]`). Returning an unbacked SymInt here should work in theory, but traceable wrapper subclass support is missing in later code to handle deferred runtime asserts. This PR fixes this.

[ghstack-poisoned]

docs/source/fx.experimental.rst

@@ -30,6 +30,7 @@ torch.fx.experimental.symbolic_shapes
     CallMethodKey
     PropagateUnbackedSymInts
     DivideByKey
+    InnerTensorKey
 
     hint_int
     is_concrete_int