Convert the PDF-embedded-image document-context item (TODO.md L3)
into phase 137_pdf_image_context (Protocol B append, 5 tasks):
per-page text extraction, the forward context path into the vision
prompt + image_context, the idempotent per-sync backfill for existing
image docs, the From-loc display line, and the dedicated E2E suite.
TODO.md cleared (items now live in .agents/phases/).
LLM-Generated: true
