feat(all): dynamic state sync

[alarso16]

This PR denotes rebased commits from two other PRs (#771 and #830) to the new clean libevm history. The first PR denotes changes to geth's snap protocol to enable usage in our repo, since a there were a few differences (the comment describes it). The second PR is the final step in which there was a successful state sync on Fuji prior to Fortuna upgrade.

There's a few things that I should mention that are mostly related to structure.

• The snap_manager.go logic is very similar to that of the downloader.go in geth. The SnapSync() method essentially processSnapSyncContent. I removed a lot of the (seemingly) excessive logic here, but it could still be simpler - specifically the stateSync channeling. There's also technically some race conditions involving the channelling in snapManager on interruption, but I'm pretty sure it will always exit gracefully with helpful error messages.
• Prior to this PR, the state trie sync and atomic trie sync could be done parallelly or in either order. Now, it is changed to completely finish the atomic trie sync before beginning the snap sync, because we will perform atomic operations synchronously throughout the state sync. However, we could likely still do these parallelly. A channel was added to notify when atomic sync is completely, and we can thus start executing atomic operations as normal.
• We cannot postpone verify/accept operations on atomic operations until the pivot, because this can cause a deadlock with the P-Chain in avalancego. Because of this, we perform atomic operations synchronously, and then after the sync is complete, we perform the non-atomic operations in order as they are buffered to catch up the state trie to the atomic trie.
• Passing the lock from stateSyncerclient to DynamicSyncer is necessary because we need to block from queueing blocks once we finish the state sync, and then once the queue is cleared, we release the lock, allowing blocks to be verified as normal.
• The DynamicSyncer was designed to use the minimal manager interface, which is exactly what is used by x/sync, but with a small type change for convenience right now (x/sync uses ids.ID for the root). The goal was to completely separate the eth-syncing logic from the DynamicSyncer, which focuses on the queuing and block operation logic.
• Currently, this requires a slightly-altered libevm repo - the branch has the same name. This is the majority of seemingly unrelated changes with the StackTrie. Without PathDB, this should be unnecessary

[darioush]

note once the new state sync algo is in place, the legacy handler can be removed and all peer requests can directly be forwarded to the "SDK" handler.

[darioush]

nit: I think you could get this like: vm.LastAcceptedBlock().Height()

[alarso16]

Hmm, this is the one tracked by the engine?