DAG Service integration

[hannahhoward]

Largely putting this one off for last, but:

The DAGService and asssociated walking functions are essentially the core traversal methods in go-IPFS

For Lightning storm to work, it needs to work with the DAG service.

My first version thought is:

• Lightning storm is essentially an indepedent library that is used to supercharge the speed of DAGService dag walks -- it's kicked off in pararel to the dag walk and hopefully writes to the local blockstore before the DAGService gets to it
• if it doesn't get ahead of the DAG walk, no worries -- they're complementary actually -- lightning storm will use the BlockService so that if blocks are already fetched by the dag service, we'll just read them from the dag store.
• we integrate this into all of the walk functions in go-merkledag
• there are probably other places walks happen in go-ipfs that we'll have to add to eventually
• this works well for backwards compatibility as once Backward compatibility for graphsync 2.1 protocol go-graphsync#376 ships we can just short circuit the lightning storm work if Graphsync 2.1 is not present on the other end

There's a penalty paid in terms of double disk I/O but I think that's probably ok until we can delete the DAGService in a much later endeavor.