v0.22.0

Added

• PDF indexing (enableData: true): .pdf files are now indexed by the data module via @firecrawl/pdf-inspector (optional dep, pure Rust, no OCR, no network). Each page becomes one row with columns page, content, needs_ocr, has_tables, has_columns. Text-based PDFs process in under 200ms locally.
• Mixed/scanned PDF handling: text pages index normally; scanned pages are flagged with needs_ocr = true and surfaced in kirograph data quality.
• Encoding issue detection: kirograph data quality warns when hasEncodingIssues is set on a PDF dataset (garbled font encodings that may require OCR pre-processing).
• kirograph data classify <file>: new subcommand — fast (~10–50ms) PDF classification without full indexing. Reports type (TextBased/Scanned/Mixed/ImageBased), confidence, page count, and which pages need OCR. Supports --json.
• PDF code-reference detection: kirograph data lint / linker now detects readFileSync, createReadStream, open, pdfplumber.open, fitz.open, and PdfReader calls referencing .pdf paths.
• metadata_json column on data_datasets: stores PDF-specific metadata (type, confidence, title, encoding issues, complex layout) surfaced in kirograph data describe.
• Platform support: prebuilt binaries for linux-x64 and macOS ARM64. Other platforms degrade gracefully (isAvailable() = false, PDFs skipped with a lint warning).
• Optional dep: npm install --save-optional @firecrawl/pdf-inspector.