feat: resharding proof

[Longarithm]

Generate proof for resharding (retain_multi_range) and check that it is valid against existing retain tests, already including memtrie and disktrie.

Surprisingly, #12390 allows to make the change almost invisible. The actual work here is to move &mut TrieRecorder inside TrieChangesTracker which is inside MemTrieUpdate.

Reasoning: when we work with MemTrieUpdate, it is in fact unique owner of the logic to record proof (what I called "2nd stage" in the previous PR, after "1st stage" of trie lookups). And now, if we are sure that proof doesn't need values, MemTrieUpdate can fully hold the &mut, because memtrie has all the necessary information, and we can record nodes directly on node access instead of doing it retroactively (which was really bad).

TrieRecorder now is passed as a separate mode to process memtrie updates. We indeed need three modes - on loading we don't record anything, for non-validators we save trie changes, for validators we also save proofs. And after that, all we need to do for retain_split_shard is to create TrieRecorder and in the end get recorded storage from it.

Next step will be to validate this proof in the actual state witness processing...