We use pretty conservative maximum lengths (255 bytes for keys, 65,535 bytes for values) for simplicity, compactness in the encoding, and to encourage efficient usage (large keys degrade performance and take up a lot of space in proofs). I could see these limits eventually changing if really necessary, but I'd like to keep them for now.

However, the limits are currently only enforced by panicking when encoding nodes - we should have earlier checks for these limits which error rather than panic.