Move the IPFS API Object to the BlockStore

[evan-forbes]

One of the requirements needed to complete #184, is that we have to retrieve block data from IPFS instead of using the PartSetHeader. To make this possible, we should move the IPFS API object from the consensus.State object, to the store.Blockstore object.

[liamsi]

Does the BlockStore really need the whole IPFS object? I think the Dag should be enough.

Also, ideally, the blockstore would only be used for local storage and retrieval. Or, at least callers should be able to indicate if they want to do network IO or only local IO somehow. We can treat the ipfs API as a blackbox for a first PR and not differentiate between the two, but we should follow up on this.

[liamsi]

Thinking about this further, the blockStore, when accessing the ipfs API in any direction reading/writing, will always potentially do some networking IO. Even when just storing, my understanding is that triggers publishing/gossiping the "have-lists". And that is probably OK. It would still be good if local vs remote would be a bit more explicit (optionally) at least when reading from the store.

[evan-forbes]

That's a great point, we definitely need to be explicit when using IPFS to store and retrieve data. This will be very important for future PRs/issues that focus on storing and retrieving data with the IPFS dag object that is being moved in #356