Support reexportability in torch.export

Summary: Title

Differential Revision: D47890879

fbshipit-source-id: 880ed5495e866a0e7ec750cba9bcc69d946191dc

Author: tugsbayasgalan

backends/test/hta_partitioner_demo.py

@@ -57,7 +57,8 @@ def forward(self, x_raw, h, c):
         input_x = torch.ones([1, 32])
         input_h = torch.ones([1, 32])
         input_c = torch.ones([1, 32])
-
+        # To avoid `operator torch._ops.aten.mkldnn_rnn_layer.default is not Aten Canonical.`
+        # error
         pattern_lstm_conv_lifted = (
             exir.capture(
                 LSTMConvPattern(),