Refactor snapshot to handle both verkle and mpt simultaneously

[gballet]

Shadow forks larger than 128 blocks will fail, because the snapshot and base tree are lost after 128 blocks. This means that the base MPT can not be read, and that the snapshot - necessary for the transition - is also unavailable.

There are several ways to proceed:

1. Iterate the MPT during the conversion. This is doable in PBSS, but a bit slow.
2. Keep more layers in RAM. This is a hack that works well, but is not going to work for mainnet.
3. Make the snapshot support pre- and post-fork values + activate the archive node during the transition, so that the MPT can be found, and the old view of the snapshot can be iterated during the conversion.

#3 is the only sustainable approach, but it requires more thinking and it probably makes sense to only use it after the rebase.