Remove the DataAvailabilityHeader from the Block

[evan-forbes]

It was discussed synchronously that we don't actually need to have the data availability header in the block. I think this makes sense, as the data root is already in the header, and tying the DAH to the block forces the implementation to include it in all of the storage/handling of the Block struct.

I could definitely be missing something, could we actually remove the data availability header from the block?

[adlerjohn]

It could be not part of the block in the same way the last commit doesn't need to be part of the block, if you define "block" as just the transactions and messages https://github.com/celestiaorg/celestia-specs/blob/ec98170398dfc6394423ee79b00b71038879e211/src/specs/data_structures.md#block.

There's no downside to decoupling other than engineering time.

[liamsi]

It could be not part of the block in the same way the last commit doesn't need to be part of the block

The last commit can't be re-computed by info that is in that same block. The DAHeader can though.

There's no downside to decoupling other than engineering time.

You'd need to store both decoupled anyways. Otherwise, when serving a DAHeader, you would always have the full cost of loading the whole Block structure from disk. The actual engineering cost was incurred when including it in the Block struct in the first place.

[evan-forbes]

this was completed a long time ago when switched to treating tendermint like a black box