Even though we no longer fully instantiate nodes before using them (allowing them to remain a NodeSprout until we are about to verify), we still key all known nodes using their checksum address.

Thus, any time many more nodes are added to known_nodes (which happens essentially on every run at some point) is thus a substantially expensive CPU operation compared to everything that happens around it. It is noticeable even on a powerful laptop in the large fleet tests. On a small device, it might be crippling. We can decrease this cost by keying using the raw bytes (known in web3 terms as "canonical address") instead.

Using checksum addresses is of course required for most transaction operations. How do other projects aimed at small devices handle quick indexing of more than a hundred or so nodes?