Revamp Multi-store Functionality

[aaronc]

This is linked to meta-issue #7096.

Summary

It was mentioned that the SDK's current multi-store is suboptimal. I believe because it uses several IAVL trees underneath. Can you share any more details @alexanderbez or @ethanfrey ?

[alexanderbez]

The main issues I see are:

1. The code is a mess.
2. Certain abstractions are either completely unnecessary or are over abstracted
3. Logical commits are not atomic

[alexanderbez]

This is something I wanted to take a look at when first mentioned. Being that we're nearly wrapped up for stargate, I'm going to start taking a look at design now...

[ethanfrey]

For 99% of use cases, with the possible exception of ethermint, we can simple have one iavl tree and then all substores are just prefixedstores on top of the iavl tree.

You get isolation and you get one tree with atomic commits.

You do not get to mix different storage formats and database backends. Do you need that for the transient stores?

[alexanderbez]

Correct @ethanfrey, I already have this in mind and will try to come up with a solution that allows for both paths.

[tac0turtle]

closing in favour of other issues