Due to padding, revising a Merkle root causes problems if the new
data is not a multiple of 64 (the segment size). For example, if two
revisions are made, each adding 32 bytes, the result is two leaves:

[32/64] + [32/64] -> Merkle root

But when it comes time to create the storage proof, the data will be
read contiguously, resulting in just one leaf:

[64/64] -> Merkle root

The easiest way to address this is to pad the revision data to the
segment size. A more robust solution would remove this restriction,
perhaps by reading into multiple Merkle trees. Fortunately, neither
of these solutions break compatibility.