Content chunks are encoded as JSON

Before storing a chunk, it gets encoded as JSON so instead of a line break, the chunk contains \n. This could lead to strange results as LLMs certainly treats actual line breaks and \n differently. The content chunks should be stored unmodified.