[Gal-Gilor]

What changed

Adds a "Cleaning text from PDF parsers" subsection to the README, under Usage between "Breadcrumbs" and "Limitations". It:

• Explains that PDF parsers produce ligatures, non-breaking spaces and fullwidth characters, that the splitter doesn't clean text, and that cleaning before split_text keeps the spans indexing into the cleaned string.
• Gives an example: unicodedata.normalize("NFKC", pdf_markdown), then removes the soft hyphen () and zero-width space () with str.replace, then calls MarkdownSplitter().split_text(clean). The two characters are written as escapes so they are visible and survive copy-paste.
• Has a table of what NFKC fixes and what it leaves alone. Accents are kept (Müller), superscripts are folded (x² becomes x2), and soft hyphens and zero-width spaces need str.replace.

No code changes, no new parameters or dependencies. No CHANGELOG entry: this is a docs-only change.

Verification

• Merged main (after Make MarkdownSplitter an immutable class and remove the outline #41, Clean up the splitter internals #43, Return only the header line from to_markdown for empty sections #44 and Test code, not the libraries, and run the doctest #45) into the branch. The diff against main is the README section only.
• Checked each table row with unicodedata.normalize("NFKC", ...).
• Ran the README's Python block as written against main's API. The sample text has fi, fl, a non-breaking space, a soft hyphen, a zero-width space, fullwidth letters and digits, and x². The cleaned text has none of the first five, and clean[start:end] for each section span starts with the header line and ends with the section's content.
• uv run pytest -q (90 passed, including the doctest) and uv run ruff check --no-fix . pass.