service/block: Implement ring buffer to cache latest X blocks for easy sampling access

[renaynay]

Question: Do we store the block data in mem for a while and then pop it out after X number of new blocks?

@mattdf 's answer: This sounds reasonable to me, perhaps do what geth does and allow the user to specify a command line argument that sets the max RAM allowed to be used by the cache, so that very big blocks don't DoS nodes in the future. If a node has to restart for whatever reason, then you'd also have to rebuild this cache by reading the disk. I think some kind of ring buffer struct would be fine here since for sampling you'll want to access the entire block.

Ref

[Wondertan]

We store chain blocks as IPFS blocks/shares rn and for caching to be implemented we can rely on IPFS caching on the share level by #108 and as implemented in #136.

By applying share-level caching we also power up caching for DASing and block syncing. Though we still may have a ring buffer for block pointers to keep them in-mem for some time.

We should elaborately consider caching for blocks here in both per block and per share levels.

[Wondertan]

We now rely on shares and we have shares being cached in #108