Transaction cache for proving backend

[bkchr]

When we calculate the storage root inside the runtime we cache the resulting transaction. This transaction can be applied to the backend to get the state of the block. After #3860 is merged, the cached transactions can be used in the import process to not require re-executing the freshly build block. When executing on the default state-machine backend the caching works and we don't need to regenerate the transaction. However, when we want to use the proving backend together with the transaction cache, the compiler currently denies it. The problem is that we can not proof to the compiler that the transaction type used by the proving backend is the same as we give it. (This leads to re-generating the transaction) To solve this, we need to clean up the state-machine Backend trait:

• This method to converting a Backend into a TrieBackend should be removed. Every backend should be a trie backend.
• The only stopping point is the InMemoryBackend that currently requires some special treatment.

This probably requires someone with a good understanding of the trie to rewrite the InMemoryBackend. I can mentor/help to some extend.

[rphmeier]

This method to converting a Backend into a TrieBackend should be removed. Every backend should be a trie backen

How even mighty abstractions fall :)

Originally the idea was to make storage totally generic so we could experiment with 2-tries, and now there are other things like RSA accumulators, verkle tries, etc.

[burdges]

Just curious: At what point do we populate the cache? If block production then this only impacts the block producer itself. If memepool then I'd hope we eventually purge all memepools from polkadoit with later sassafras phases, although that's quite a ways off.

[bkchr]

We populate it while building blocks.