Previously, this would measure as <20 bytes for addresses with leading
zeroes. This resulted in transactions with unexpected behavior.
For example, sending ETH to the zero address would create a contract
instead.
Here we switch to using +encode:rlp directly and indicate a width of
20 bytes for the "to" field.
Null bytes in lists were getting eaten during concatenation. To avoid
this, we track encoded item widths (which are always 1 or higher) and
+can them all together.
This likely did not affect any of the other Ethereum code, considering
it nearly always measures atoms, and the null byte would be seen as
no bytes in that case.
