perf: cache PDFium document across page operations

[AdemBoukhris457]

Summary

• Add loadDocument()/closeDocument() lifecycle to PdfiumRenderer so the PDF is loaded once and reused across all per-page calls instead of re-parsing on every invocation.
• Wire up caching in PdfJsEngine (extractPage + renderPageImage) and LiteParse.screenshot().
• Maintain backward compatibility: methods fall back to per-call load/destroy when no document is pre-loaded.

Problem

Both renderPageToBuffer() and extractImageBounds() in PdfiumRenderer perform fs.readFile() + loadDocument() + destroy() on every single call. For a 100-page document this means 100-200 full PDF load/parse/destroy cycles with redundant disk I/O, CPU usage, and memory churn.

Changes

• src/engines/pdf/pdfium-renderer.ts: Add cachedDocument field, loadDocument(), closeDocument(), and private getOrLoadDocument() helper. Refactor renderPageToBuffer() and extractImageBounds() to reuse cached document. Update close() to clean up cached document.
• src/engines/pdf/pdfjs.ts: Call loadDocument() on the PdfiumRenderer after creation in extractPage() and renderPageImage().
• src/core/parser.ts: Call renderer.loadDocument() in screenshot() before the page loop.
• src/engines/pdf/pdfium-renderer.test.ts: Rewrite mocks for repeated calls, add 4 tests for caching behavior.
• src/engines/pdf/pdfjs.test.ts, src/core/parser.test.ts: Add loadDocument/closeDocument to PdfiumRenderer mocks.

Closes #88