Investigate if store functions should return a `Result`

[plafer]

We currently return early if one of the writes to the store fails (e.g. here). ibc-go assumes writes never fail (e.g. here).

We should think whether we want to also assume that writes never fail, or if not, what to do when a write fails.

[romac]

Any store backed by some kind of I/O device (file system, network, memory in some specific cases, etc) can and likely will fail at some point. Therefore I believe we should reflect this at the type level and deal with those failures accordingly (eg. by simply bubbling them up to the host application).

What do you think?

[plafer]

Right, I thought some more about this since yesterday, and I think it's easily solvable if we document that "if validate/execute/deliver() returns an error (any top-level function), all state changes need to be reverted". This is what we do in basecoin-rs, and I think an undocumented assumption that we make everywhere in ibc-rs (which makes sense but I still keep forgetting from time to time). The fix is probably to just clearly document this requirement.