Implement MultiStore with fallback mechanism

Based on discussion and the work done in the PoC,

We're going to be reimplementing a MultiStore that looks like a redis-rb Redis instance (which the Redis::Store inherits from). We're going to be reimplementing reads and writes separately so we can have a read-one-fallback/write-both state. It's easy to miss a method so we'll method_missing it to the SharedState Redis and keep track of it to add an implementation. This ensures that anything unexpected behaves as it would before.

We should also want to add metrics in order to keep track of how many times we needed to fall back when we tried to read from the shared_state(fallback) instance.

We're paying special attention to pipelined and multi methods, so the blocks within them talk to the right Redis instance and not both.

The MultiStore should talk directly/only to SharedState by default. We can enable read-one-fallback and write-both with a feature flag.

This is part of the preparation work for our Sessions: Plan migration strategy