The standard methodology is to associate a single de Bruijn graph node with both a k-mer and its reverse complement. If you take a look at e.g. GATB's excellent implementation, you'll see that this is how they handle it. The problem is that when processing sequencing reads, you really have no idea what strand a read was drawn from, and so you don't really have a good way to distinguish between the two. It's also worth noting that one typically builds de Bruijn graphs with only odd-sized k-mers. This prevents the scenario where a k-mer is its own reverse-complement.