[storage + vm] Allow VM to delete proposervm blocks (or make blocks smaller)

[patrick-ogrady]

Allowing a VM to delete old blocks from the proposervm (and from its block store) would prevent ever-growing disk size (relying entirely on state sync for new nodes to join the network).

On a tokenvm devnet with 10k accounts that send 100M transactions, only 125MB of state is used but 32 GB of block storage (16 GB in tokenvm and 16 GB in proposervm) is used. The state size remains constant during this test but block storage grows unbounded.

[patrick-ogrady]

We can avoid this problem if we only put hashes of transaction blobs into consensus (instead of the full block) and then prune transaction blobs after some period of time.

This also allows us to have blocks that contain more than 2MB of data.

We would fetch these blobs during parse concurrently over the network layer.

One interesting observation to consider would be attempting to use the same tx blob of other processing blocks at the same height.

Perhaps a more interesting observation would be to pre-gossip the tx blob to other nodes (where nodes would validate if they got that message from someone producing soon -> maybe with a signature to attest to this to avoid having producer send to all).

We could alternatively have nodes gossip the tx blobs they possess (ordered from newest to oldest to try and optimize the number of O(1) fetches). This should be of the format <oldest height, accepted height, [processing roots held]>.

[patrick-ogrady]

At some point, the only way to continue increasing TPS is to amortize the state root generation over larger numbers of txs (that hopefully update some components of the same state).

[github-actions]

This issue has become stale because it has been open 60 days with no activity. Adding the lifecycle/frozen label will exempt this issue from future lifecycle events.

[patrick-ogrady]

Blocked: ava-labs/avalanchego#1929

[patrick-ogrady]

Add an optional StoreBlocks config that stores block by height for later query. It would also be nice to expose a DeleteBlock API so that indexers can remove blocks after they've been indexed. This storage mechansim should also store results (for those not using the WS index).

We should move this into the VM scope to remove duplicate, useful code from MorpheusVM/TokenVM.