Stratification should flatten names after each application instead of using nested tuples

Currently, multi-level stratification returns state and transition names as nested tuples. This becomes an issue when serializing/deserializing.
- When deserializing, the JSON strings must be converted to tuples of symbols, but calling `flatten` on a tuple of tuples of strings and singleton strings will separate the individual letters of the singleton strings resulting in the incorrect names: e.g.,`Symbol.(flatten((("guns", "n"), "roses")))`.
- Trying to avoid the issue by flattening prior to serialization also does not work because, if `flatten` is called on a tuple of tuples of symbols and singleton symbols, the iterator will not be able to iterate over the singleton symbol:
e.g., `[x for x in flatten(((Symbol(1), Symbol(2)), Symbol(3)))]` will error.

Flattening the names after each hierarchical stratification, s.t. the names are always a non-nested tuple, would avoid the issue. It also makes sense categorically because the product models are defined up to isomorphism, so the associations indicated by the nested-tuple names are not meaningful.