This PR speeds up of vm::LargeBocSerializer by implementing batch fetching with RocksDB::MultiGet instead of sequential RocksDB::Get operations. Key changes include:

• Replaced depth-first search (DFS) traversal with breadth-first search (BFS)
• Implemented level-by-level cell fetching using batched MultiGet, with max batch size of 4 mln.
• This implementation loads batch size cells in memory, which leads to increased memory usage (with batch size 4mln - at peak 15-20gb)
• Since in-memory implementation vm::BagOfCells still uses DFS, the cell order in resulting BoC differs from result of vm::LargeBocSerializer. Therefore test that performs BoC byte to byte comparison in test-db.cpp was disabled.

This change speeds up state serialization up to x3 times.