Describe empty data edge case

[alpe]

Updates doc to describe the edge case for response overwrite in submessages

Follow up from: CosmWasm/wasmd#534 (comment)

[webmaster128]

With this solution, there is no way for a contract to clear previous data. E.g.

// Execute
data: Some(b"first thought")

// Reply
data: Some(b"") // Okay, better shut my mouth

Do we really want to do that? Or better increase testing for the null/empty differentiation?

[ethanfrey]

I cannot conceive of such a case, but I do admit there are cases we thought would never come up, which later do.

That said, the whole Some(b"") vs None his us in many places, and I am still not convinced that in any of them we really needed to differentiate them.

Where we pass these values in JSON, we need to explicitly use custom types in Go, as the default parser for []byte doesn't distinguish between these two "zero values".

[webmaster128]

Where we pass these values in JSON, we need to explicitly use custom types in Go, as the default parser for []byte doesn't distinguish between these two "zero values".

I think we tested that and it worked well. It just collapses with strings as the string type in Go is not nullable but []byte is.

[ethanfrey]

Hmmm... you are right... This does produce different values for null/missing and ""
https://play.golang.org/p/KqI86DXQ7jq

I guess it makes sense to test this and enforce different behaviors.

I was concerned about distinguishing them and advised Alex to be careful about handling these two cases differently