Wire `Documents.add_pdf()` / `Documents.add_csv()` into the public client API

[cferrys]

Summary

The Python client currently advertises first-class local PDF and CSV ingestion, but the public Documents API does not actually implement add_pdf() or add_csv(). As shipped, the documented examples will fail at runtime with AttributeError, leaving a half-integrated ingestion feature in the newest modular client.

Context

The recent modular client work added parser utilities for local files in utils/pdf_parser.py and utils/csv_parser.py, and the README/examples position them as supported ingestion workflows. However, lib/documents.py only exposes CRUD, import, filter delete, and search methods; there is no public bridge from the client to those parsers.

This creates a high-impact mismatch in a search engine client, because document ingestion is the first mile of the indexing path. A broken ingestion helper blocks evaluation of the engine on real PDFs/CSVs and undermines trust in the client surface right after the new package/API expansion.

Relevant locations:

• lib/documents.py
• utils/pdf_parser.py
• utils/csv_parser.py
• README.md
• examples/pdf.py
• examples/csv.py

Proposed Implementation

1. Add Documents.add_pdf(collection_name, file_path, options=None) and Documents.add_csv(collection_name, file_path, options=None) to lib/documents.py.
2. In each method, parse the local file via the existing helper, build the document payload, and submit it through the existing add() path so validation/auth/request behavior stays centralized.
3. Normalize option handling so callers can override document fields, IDs, metadata inclusion, and parser settings without bypassing validation.
4. Add tests covering:
   • successful PDF/CSV ingestion helper flow
   • missing file / missing optional dependency failures
   • generated document shape
   • README example parity
5. Keep README/examples aligned with the actual shipped API surface.

Impact

This fixes a user-facing runtime bug in the documented happy path, completes an already-started feature, and makes the Python client materially more useful for real indexing workloads. It also reduces onboarding friction for a high-performance search engine, where fast ingestion from local datasets is often the first proof-of-value step.