Put every ~10k block hashes in the state tree

We could put every 10,000th block hash in the state tree. This doesnt add too much overhead to the chain (~4kb/year + overhead) and gives us ~constant sizes merkleproofs for any block in the chain for light clients. In the even more future, we could have snark proofs that one block is the correct descendant of the one 10,000 blocks before it (snarking the state transitions). Once we have that, all of chain sync becomes constant time (roughly).

Start by reviewing the state tree and chain-sync parts of the Filecoin protocol specification, then trace how block hashes and Merkle proofs are represented for light clients. The work is complete when every 10,000th block hash is included in the state tree and the resulting proofs have roughly constant size, with overhead considered.

Written by the indexing model from the issue text.