Do Node filtering before interneurons

This makes them like Ensembles in this respect, and should avoid
problems with different taus on multiple inputs to an ensemble.

Fixes #94.

nengo_loihi/splitter.py

@@ -180,7 +180,8 @@ def split(model, loihi_model):  # noqa: C901
                 with chip:
                     logger.debug("Creating ChipReceiveNode for %s", c)
                     receive = ChipReceiveNode(dim * 2, size_out=dim)
-                    nengo.Connection(receive, c.post, synapse=c.synapse)
+                    nengo.Connection(receive, c.post,
+                                     synapse=loihi_model.inter_tau)
                 with host:
                     max_rate = loihi_model.inter_rate * loihi_model.inter_n
                     rtol = 1e-8  # allow for floating point inaccuracies
@@ -210,7 +211,7 @@ def split(model, loihi_model):  # noqa: C901
                                      solver=c.solver,
                                      eval_points=c.eval_points,
                                      scale_eval_points=c.scale_eval_points,
-                                     synapse=None,
+                                     synapse=c.synapse,
                                      transform=c.transform * scaling)
                     nengo.Connection(ens.neurons, send, synapse=None)
                 host2chip_senders[send] = receive