decompose class hierarchy of LLM implementations

[Ron S]

decompose class hierarchy of LLM implementations:
from one class per provider → one class per model and it's respective capabilities

Current problem:
you may call functions on a model that the model does not support, as there is only one class per provider that all models use
in particular, ChatWithFunctions extends Chat, whereby ChatWithFunctions is not necessarily compatible with Chat, thus modelled incorrectly which can lead to unsupported operations at runtime