Potential race condition in flow setup

[vaterlaus]

When SNAC gets a packet_in, it responds with two messages.

1. a flow_mod entry to change the flow table
2. a packet_out message with actions output:TABLE to free the buffered packet_in packet
   The issue is that message (1) is not guaranteed to be finished processing before message (2) is processed, and it is likely that the first packet in a flow is lost, which can cause annoying delays for users.
   We have not seen this bug often in practice because most switches are single threaded. That said, it is likely an issue that should be resolved.

Reported by: capveg
Priority: minor