Deliver an intermediate data format as AI input

[rschwietzke]

AI, especially LLMs, do not deal well with our large XML intermediate data. Let's provide a second LLM-ready data set that might also be reduced (no error message, no counts, for instance). There might not be one single format, but maybe two, just as JSON or Markdown. And all numbers might also not be required.

Let's play the concept in a way that we can define a render template in the project, so this can be adjusted if needed; otherwise, it will fall back to a good default provided by us.