Plural rule parsing and handling code cleanup

[echeran]

In #1091, we recently discussed and concluded next steps that are specific to the immediate short-term needs of the imminent v1.0 release. In the process of working on #2598 to implement those specific changes, some difficulties arose due to the code needing some cleanup.

Issue(s):

• One of the original tasks is to make icu_plurals::rules experimental. However, this is not fully possible
• The existing submodules icu_plurals::rules::reference and icu_plurals::rules::runtime had to be marked as #[doc(hidden)] because they could not be put behind the "experimental" attribute
• The reason that icu_plurals::rules::{reference, runtime} could not be marked as "experimental" is because runtime has serde implementations for JSON, and runtime depends on reference

Instead, what we want is code that we discussed as experimental code -- not currently intended for stable API and intended for further discussion -- to be properly put behind the "experimental" attribute accordingly.

@zbraniecki

[sffc]

We should either move the code to its own crate, or establish a new feature in icu4x that means "this code is not stable but also not experimental".

It's a bit tricky that runtime depends on reference; it means we can't just block out the whole reference module from compilation. We may need to compile it all the time, but have a conditional re-export of it that we can put in a feature.

[zbraniecki]

We could reverse the order and have reference reference runtime but then extracting it to a separate crate will be harder.

[sffc]

The dependency is deeper than that; runtime needs reference because we use reference for serializing into JSON.

[sffc]

The dependency is deeper than that; runtime needs reference because we use reference for serializing into JSON.