Cookbook: bank statement extraction with cross-field validation

[pranjalparmar]

This is the best showcase for the project's most distinctive feature, the failed_check flag from a custom cross-field rule (e.g. line items summing to a stated total), and it currently has zero worked example anywhere in the docs.

What's needed

A new docs/cookbook/bank-statement.md (create docs/cookbook/ if it doesn't exist yet) containing:

1. A schema for a bank statement with a list of transactions (date, description, amount) plus a stated closing balance or total.
2. A custom cross-check rule using numeric_sum_rule (already exported from fastdocparse, see grounding.py) or a hand-written rule, wired up via DocumentParser's cross-check hook (see docs/output-format.md for how custom rules plug in).
3. A real synthetic bank statement (do not use a real person's actual financial document) run through fastdocparse extract, with the cross-check rule enabled.
4. The actual output showing a case where the check passes, and ideally a second run showing it catching a real mismatch (e.g. deliberately altering one transaction amount in the source document), flagged failed_check. Both need to be genuine output from a real run, not fabricated.

This requires an actual LLM call (OpenAI key or local Ollama model), that's why it's a separate task rather than something scripted.

Where it's linked from

Add a link to the new page from the "Full documentation" list in README.md.