Implement Block Data propagation about celestia-core

[liamsi]

The way we (roughly) want to approach this for a very first version, is the following:

The block proposer, will, after erasure coding the block into a square and compute the row and column roots:

We need to write the nodes of these trees into an IPFS DAG. To make this possible, we first need to collect the nodes of the Merkle trees - the current NMT implementation returns only the root of a tree and throws away the inner nodes (see: celestiaorg/nmt#9).
Then, we need to implement a IPLD plugin, similar to the one used in our experiments to properly store the nodes into the ipfs DAG.
To be useful this requires running an IPFS node along with a tendermint node.
There are two options how we could write into the DAG:

1. Use dag put: This requires the nodes to be encoded into a format the plugin can parse prior storing them.
2. Directly batch add the nodes: dag put internally decodes and then batch adds the decoded nodes via the DAG api / a NodeAdder. We probably can directly use this adder instead.

(1.) has the advantage that we can also use this via the command-line and we know that it works. The only downside is that it requires a, in this context, superfluous encoding and decoding step (roughly Nodes->Encode(Nodes)->Decode(Nodes)->Nodes->AddBatch(Nodes))
(2.) does directly what we want. We could even pass a "store" into the NMT, that adds the nodes directly to the batch. (roughly Nodes->AddBatch(Nodes)). Downside (not really) is that this might require some more digging into how to setup a environment.

As a first step this block propagation mechanism will be added to the current gossip-based mechanism, later we will remove the current mechanism.

[liamsi]

We decided to stick with Tendermint's block propagation for now. closing.