[WIP] make retriever page content just the content

[lgabs]

Addresses this dialog issue. This retriever is only used in lcel.py for now, for instance:

https://github.com/talkdai/dialog/blob/14e19f619155a710de4c350c0a78adc1942ae9d6/src/dialog/llm/agents/lcel.py#L60-L61

Users that already uses this project as-is, leaving questions in the question field and answers in the content csv field would have to be informed that this change will demand them to either join question/answer before into the content field or chose another format_docs function (which will be coerced into a runnable), in this case by using a plugin.

[vmesel]

@lgabs coding wise it makes sense, but on the optimization side: don't we need the title/question for the content to be more relevant? Or is it added to the context as well?

[lgabs]

Yes, the title and question are relevant together, but my point about flexibility is that the retrieval phase should only focus on the task of returning the content and their metadata (in which case title is a metadata). Then, when combining documents later in the chain, users are free to combine as they wish, for instance the following prompt template strings could all be valid prompt (compared to that example in [dialog])(https://github.com/talkdai/dialog/blob/main/src/dialog/llm/agents/default.py#L53):

"Question: {question}\nAnswer{content}"
"Question: {question}\nCategory:{category}\nAnswer{content}"
"Question: {question}\nCategory:{category}\nSubcategory:{subcategory}\nAnswer{content}"

[vmesel]

Closing this PR until we have any news here.