Add pypdfium2 as optional PDF parser

[rejojer]

Summary

• Adds opt-in pdf_parser="pypdfium2" alongside the existing PyPDF2 default
• Default behavior is unchanged — existing users see no diff; pypdfium2 is lazy-imported only when selected
• pypdfium2 yields cleaner text (no Reser ve-style broken words, correct Unicode like †/‡) and is roughly 3-5x faster on the corpus in examples/documents/

Changes

• utils.py: new read_pdf_pages(doc, pdf_parser) helper with PyPDF2 / pypdfium2 / PyMuPDF branches; get_page_tokens now a thin wrapper over it
• client.py and retrieve.py: route their direct PyPDF2.PdfReader calls through read_pdf_pages so the parser choice flows end-to-end (parser is recorded on the document so cache-miss reads stay consistent)
• page_index() and PageIndexClient.__init__() accept pdf_parser parameter (default None, falls back to config.yaml)
• config.yaml: pdf_parser: \"PyPDF2\" default
• run_pageindex.py: --pdf-parser CLI arg
• requirements.txt: pypdfium2 noted as optional (no new hard dependency)

Usage

# default unchanged
client = PageIndexClient()

# opt-in pypdfium2 (after `pip install pypdfium2`)
client = PageIndexClient(pdf_parser=\"pypdfium2\")

python run_pageindex.py --pdf_path doc.pdf --pdf-parser pypdfium2

Test plan

• ConfigLoader().load().pdf_parser == \"PyPDF2\" (default preserved)
• read_pdf_pages(pdf) byte-identical to PyPDF2.PdfReader(pdf).pages[*].extract_text()
• get_page_tokens(pdf) with no parser arg produces identical output to pre-PR behavior
• pdf_parser=\"pypdfium2\" extracts pages, normalizes \r\n to \n, accepts both file paths and BytesIO
• PyMuPDF branch still works
• Unknown parser name raises clear ValueError
• Missing pypdfium2 raises ImportError with pip install pypdfium2 hint
• page_index() and PageIndexClient.__init__() signatures include pdf_parser