Function normalize_text 

pub fn normalize_text(raw: &str) -> String

Canonicalize extracted text so output is stable across adapters:

1. Normalize line endings to \n (drop \r).
2. Trim trailing whitespace on each line.
3. Collapse three-or-more consecutive blank lines to a single blank line.
4. Trim leading/trailing blank lines, then append exactly one \n (unless the whole text is empty, which stays empty — the image-only-PDF contract).

This is layout tid-up only; it never reorders or drops words. Word-level content is whatever the adapter recovered.