fix(core): preserve OCR continuity for blank pages

[jb-thery]

Summary

• accept a valid empty Tesseract payload for image-only PDF pages with no detectable text
• preserve strict page-count validation for malformed OCR output
• cover the real blank-page regression with a focused test

Upgrade continuity

This prevents one blank scanned page from stopping a full reindex after an upgrade. Existing sources remain untouched, and the index can complete without excluding the document.

Validation

• pnpm validate
• 497 Core tests passed
• Landing, Chat, and TTS test suites passed
• build, security audit, smoke, packaging, and release artifact checks passed