Feat: Add support for rendering records as LLM documents

[aaronsteers]

Related to:

• Add PyAirbyte RAG example using LlamaIndex #69
• Add PyAirbyte RAG example using LangChain #70

Here we add a new Document class which is modeled after the LangChain Document class and should be compatible with similar LLM paradigms. The paradigm here is that content (or page_content) is a stringification of the record as a document, and metadata is a collection of all supporting fields that do not make up the primary content.

Later, we can add top-level id and last_updated fields, which will help us later with deduplication.

The new interfaces are:

• Source.get_documents(stream_name, ...) - this is sibling to the existing method Source.get_records(stream_name, ...). Instead of returning an iterable of record dictionaries, we return an iterable of document objects.
• Dataset.to_documents(...) - this is sibling to the existing methods: Dataset.to_sql_table() and Dataset.to_pandas().

For comparison, please see the Langchain Documents class.

Note:

• The long-term vision here would be to let sources define how to render documents within their catalog json, and then AirbyteLib can use that as a guide (when available) instead of the generic implementation. For more information, see:
  • Let sources define their own default document rendering processes #74