libevm phase 2: key decisions

[darioush]

1. Triedb backend flexibility. Goals:

• Allow coreth to specify its own triedb as backend interfaces.
• Minimally invasive for existing unit tests that are referencing hashdb.Config & pathdb.Config
• Basic implementation: ava-labs/libevm@1910a11

Alternative: Use upstream backends & make our modifications there (such as ReferenceRoot). Not sure how pathdb will be connected in this case, as we have some modifications even though it is not for use in production the unit tests are expected to pass.

2. Snapshot compatibility: In coreth & subnet-evm snapshots layers identified by the hash of the block not the root of the block.
   The approach in the draft PR keeps the interface the same in the libevm code (and handles nil ptrs vs. interfaces, need to see if/why this part is needed, and if so what solution can we do).

Possible approaches:

• The approach in the draft PR (or similar ideas), however this is not great because it needs some method called before the other one is allowed to be called. https://github.com/ava-labs/coreth/pull/669/files#diff-53d5f4b8a536ec2a8c8c92bf70b8268f1d77ad77e9f316e6f68a2bcae5303215R1696-R1704
• Try to use state roots the same as upstream. (need to handle cases with repeated state roots in verification tree)
• ? any other ideas

3. Trie prefetcher. https://github.com/ava-labs/coreth/pull/666/files shows an alternative approach that involves minimal changes to upstream code. This concurrency is a hard performance requirement. My suggestion is to review, agree on the code structure, discuss necessary benchmarks, and ship this part vs. master independently.

[darioush]

Issue 1 is resolved by ava-labs/libevm@594abd9