Rework approximate markdown servlet

[stoerr]

The algorithm used to extract the text from a page or component as input for ChatGPT currently works with a heuristic that renders common attributes like jcr:title, title, subtitle, linkTitle, jcr:description, text, code, copyright, defaultValue, exampleCode, suffix, exampleResult, footer (see ApproximateMarkdownServiceImpl.java), and add special cases for special components using ApproximateMarkdownServicePlugin . That seems to work fine for standard components, but will output nothing or little for custom components that have custom attributes (which is pretty common in AEM / Composum development). Thus we have to improve that.

Variants:

• Turn it around and output common attributes like jcr:title, title, subtitle, linkTitle, text, header, footer as they are, filter out system attributes by name and value (we probably won't want to output any attribute with a namespace and also not attributes with numbers, booleans, dates (?), arrays (?) )
• Extract the text content from the HTML rendering of a page or component. (advantage: that contains content included from elsewhere, disadvantage: it includes irrelevant content like navigations, advertisements, headers, footers etc.)

Additionally, see whether we can easily rework it to produce HTML when needed.