Cache the decoded structs instead of the encoded bytes

[cwgoes]

Ref #2187 (comment)

At the moment, I expect there's a lot of redundant encoding/decoding through the cache layer - we update a validator struct, Amino-encode the updated struct, save it to the cache, promptly read it back from the cache and Amino-decode it again! Looking at the first version of benchmarks, it looks like Amino is a substantial bottleneck; I think if we can find an elegant way to do this it might be quite helpful.

Probably not prelaunch unless it's really easy.

cc @ValarDragon thoughts?

[ValarDragon]

Lol we had the same exact though here. I'm in favor of keeping the conversation on the other issue due to having already started the conversation there, and there being a proposal to evaluate.

I think if there is anything we want to optimize, benchmarks indicate that this is the first thing we should be doing for speed.

[alexanderbez]

++ seems like we can close this one?

[cwgoes]

Closed in favor of #2193.