Use object for df and lake config instead of a dict

[gventuri]

I'm opening this conversation based on the comments in this PR: #551.

@regmibijay mentioned:

• I meant to add that each of SmartDatalake and SmartDataframe could have each set of configuration classes, which could share same Abstract Base Class. This could also make it possible to implement methods such as SmartBaseConfiguration.from_dict() which supports Kwargs. Bit like this config class for transformers.
• The main problem I see with passing kwargs is that it feels a bit hacky and there is no way of knowing which options are available. With classes and docstrings, it is possible to either view Configuration signature in IDE or with help or in case of Juypter Notebooks with SmartDFConfig?
• With 3rd point, I meant perhaps it could be easier to aggregate basic sets of parameters and config options for all LLMs. This will in return make writing tests bit easier in some cases.

So what you suggest is:

df = SmartDataframe(df, config={"llm": llm})
# replace with
df = SmartDataframe(df, config=Config(llm=llm)}

Is that correct?

At the moment, we support both the implementations and I'm not sure we should drop one or another, but I'm open to any feedbacks from you. That said I can see your point in suggesting to make the second option as default, as it can benefit from a more robust validation.

As of your points:

• I don't know if at the moment there are enough differences between SmartDataframe and SmartDataframe to justify a diffrent configuration file for each. But things might change in the future, and it might make sense to have several configuration extending an abstract one.
• Nothing to add about the second point
• Yeah, 100%. I think instead of acting at the config level, we should also do some changes to the LLM wrappers themselves, making sure that 1) they don't contain business logic apart from connecting to the LLM 2) they have a very similar structure and they accept the same params

Curios to hear your thoughts!