Check URLs in LLM output text

[jhrozek]

Especially with features like citations where the LLM might print a URL that the user then navigates to, we should add a pipeline step that checks those. I can think of two useful steps:

• check for malicious URLs using something like VirusTotal
• a configurable blocker that would just have a bunch of regexes

[lukehinds]

Makes sense to highlight #929