Post-train quant: Special handling for bidirectional DistillerLSTM (#337)

* For some reason, SummaryGraph generation is broken for DistillerLSTM
  modules with 'bidirectional' enabled. The ONNX graph optimization stage
  causes all the nodes from the bidirectional module to vanish from the
  graph (they're in the graph after the initial trace)
* As a temporary workaround to enable stats fusion in post-train quant,
  if a bidirectional DistillerLSTM is detected, we just do a simple
  "hard-coded" fusion of the element-wise add op with the subsequent
  non-linearities and skip the automatic flow with SummaryGraph.