Move to height->header index

[cristaloleg]

In most cases we don't need an index hash->header but rather height->header.

[Wondertan]

One thing I recalled is that this might change with backward sync. We may want to request a range of headers with hash + number of headers before. This way, as we know the hash is valid, we can simply verify the validity of the whole range in a stateless manner. Requesting everything with heights, on the other hand, would require external state.

[renaynay]

@Wondertan I'm assuming the hash you're discussing here is the TrustedHead (either some trusted peers gave it to us or we found through "social discovery") and we pass it to the node. We could also use the height here and pass in the hash to ensure the height we've requested == the hash we trust, and then work our way backwards (where all the reference information we need is in the headers we've already processed and verified (as header contains LastBlockID)

[Wondertan]

That's true, statelessness can be achieved solely with interfaces. The hash then doesn't even need to be part of the protocol, as it is right now.

[Wondertan]

One more thing to consider. We kept hash addressability in Shwap's storage. @adlerjohn , @walldiss and I couldnt find a compelling usecase for it but still decided to keep around as more like a standard to expose that. I would remove it everywhere after we know for sure its useless, assuming it brings unnoticeable maintanance overhead.