sp-state-machine: Read value from backend when writing the first time

[bkchr]

This is a first step into fixing #6020. The underlying problem for parachains is that when calling storage_root, values that are written, will need to be looked up in the trie. This is required to insert the value or to remove it. The problem is that this lookup in storage_root increases the storage proof size, but storage reclaim for example can not track these lookups. This means that with storage reclaim it is possible to include more transactions than what the block should be allowed to include.
This pull request reads a key from the backend the first time it is written. This results in taking into account what storage_root is doing at the end of a block. However, this may still skips reading one extra node (depending on the trie structure). Skipping only one node is far less problematic than skipping the entire lookup. A future pull request will improve this further to take the entire operation of storage_root into account.

For chains that are not recording a storage proof, they will still do this extra backend read. However, this read will end up in the cache. At the end of the block the key will be read again as part of storage_root, but served from the cache. So, there is no downside of this.

[skunert]

Won't this break backwards compatibility? If storage reclaim now "sees" more read acesses than before, it will arrive at a different reclaim amount. All chains that have the reclaim enabled would arrive at a different state (i.e. BlockWeight) after importing old blocks with this change. Even though the final proof should be the same.

Edit: Same with the PVF, producing nodes would arrive at a different BlockWeight while the state machine in the PVF would use the old logic, failing to validate new blocks.