Quarkus Langchanin4j Chat model classes need to be registered for reflection

[jamesnetherton]

I am a bit confused about this. With Quarkus Langchain 1.0.0, I find that langchain4j-chat native tests are failing. It's due to the various model types not being able to be marshalled / unmarshalled.

In the case of Ollama, I had to register a bunch of classes for reflection:

dev.langchain4j.model.ollama.FormatSerializer
dev.langchain4j.model.ollama.Message
dev.langchain4j.model.ollama.OllamaChatRequest
dev.langchain4j.model.ollama.OllamaChatResponse
dev.langchain4j.model.ollama.Options
dev.langchain4j.model.ollama.Role
dev.langchain4j.model.ollama.Tool
dev.langchain4j.model.ollama.ToolCall

Previously this was not required. We should dig into why it's needed because ideally, we don't want to manage this in CQ. Especially for all of the different Langchain4j supported models users may want to work with.