FIX: Sources from LLM

[Mookse]

OpenAI GPT engines return not only source references but sources. For phase one, given the non-sensitivity of our public content (yes, that's tautological), we can just remove them. Phase two, we'll need to honor and sophisticate.
Worth noting that this particular hallucination was able to draw up its own hallucinatory source. Although, oof, as I look at it, linguistically, the trick is ingenius. While referencing a sane object in the system (a biographer bot), and by wrapping Walter in quotes, the poor dumb kid Q never had a chance. That said, I bet it can hallucinate sources, but these may not be they.

In reality, there are no walters in the system, nor would Q have access to that information (at least at this stage, Oy future KM!), so it's not a security issue in this instance, but a fine illustration of how to test!!!

[Mookse]

• content removed

    const rSource = /【.*?\】/gs
    const rLines = /\n{2,}/g

[Mookse]

ready for phase 2 from description