Measure impact of changing ZeroVecError to start at 1

[sffc]

@kupiakos found that making error enums start at 1 can help code size:

https://twitter.com/kupiakos/status/1587506293996036097

We have a lot of cases of Result<(), ZeroVecError> throughout the code base. We should try to apply this optimization to ZeroVecError and measure the impact.

[kupiakos]

Also consider the code size impact of switching ZeroVecError to #[repr(usize)]. Rarely are errors actually stored in-line in memory, and some platforms treat word-sized arguments more efficiently than smaller-than-word-sized arguments (most notably, RISC-V).