Wrong impl of `KeyLenOf` for `Map` and `DoubleMap`

[mustermeiszer]

If I am not mistaken the length of the Patricia-Key for a storage object is determined by

• StorageValue = Twox_128("PALLET_NAME") + Twox_128("STORAGE_NAME")
• StorageMap = Twox_128("PALLET_NAME") + Twox_128("STORAGE_NAME") + HASHER(key)
• StorageDoubleMap = Twox_128("PALLET_NAME") + Twox_128("STORAGE_NAME") + HASHER(key_1) + HASHER(key_2)

But the current implementation of KeyLenOf for both StorageMap and StorageDoubleMap ignores this. It returns a key length derived from the actual BYTES of the names of pallet and storage elements, ignoring the constant size hash length.

• Map impl of Get<u32>: https://github.com/paritytech/substrate/blob/ef890f5598e0d3fd1b6a5fcc8c3322f7570089d1/frame/support/src/storage/types/map.rs#L65
• DoubleMap impl of Get<u32>: https://github.com/paritytech/substrate/blob/ef890f5598e0d3fd1b6a5fcc8c3322f7570089d1/frame/support/src/storage/types/double_map.rs#L94-L98

This, in my understanding, leads to a wrong assumption about the maximal trie depth and subsequently a wrong assumption about the maximal amount of elements in the maps.