feat(scrape): vision OCR of image-only doc pages (support matrices)

Some HPE pages carry information only inside an image. The Morpheus
release schedule / support-lifecycle matrix (sf000111242en_us) is the
case that surfaced this: the entire "which version ships when, when a
stream hits Maintenance/EOL" table is a JPEG, so html_to_md dropped
every value and retrieval could never answer "when does 8.x.x reach
end of life".

New scrape/vision.py (opt-in via VISION_OCR) transcribes qualifying
images with a local Ollama vision model and appends a labeled markdown
block to the page so the text is chunked, embedded and retrieved.

Model/prompt chosen against ground truth on the matrix (14x4 cells):
- qwen2.5vl:7b + a bare "transcribe the table" prompt: 55-56/56, ~18s,
  GPU-resident, and prompt-ROBUST (accurate without hand-tuning).
- gemma3:12b needs an exact "read row by row, keep cells aligned"
  prompt or it shifts a column by one row; qwen2.5vl:32b is no more
  accurate and 5x slower (CPU spill). Both documented in the code.
- Note: self-consistency (2 samples must agree) only catches RANDOM
  flakiness — a wrong read is stable across seeds — so every block also
  ships a "verify against the source image" caveat.

Reliability/cost:
- content-hash cache in corpus/.vision-cache/ (committed) — each unique
  image OCR'd once ever, shared across version bundles.
- VISION_MAX_NEW bounds NEW OCRs per run so the first pass can't balloon
  into hours; the cache fills incrementally. Deferred count is logged.
- every failure path degrades to the pre-vision behavior; never blocks
  the scrape.

Also:
- add the morpheus_release_schedule bundle (sf single-doc) + 2 eval
  golden queries for it.
- fetch_single_doc: title falls back to the bundle title (not docId)
  when a page has no <h1> (sf solution articles).
- Pillow in requirements-vision.txt (scrape-only; kept out of the
  server image), installed + VISION_* wired into refresh.yml.

Verified locally: matrix transcribes to the exact 14-row table; second
run is a cache hit (ocr=0).

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01LFowQzJu7k97QLCRDSAeh1

.gitea/workflows/refresh.yml

@@ -37,6 +37,18 @@ env:
   OLLAMA_URLS: http://192.168.0.2:11435,http://192.168.0.2:11436,http://192.168.0.125:11434,http://192.168.0.126:11434
   EMBED_MODEL: nomic-embed-text
 
+  # Vision OCR of image-only pages (support/lifecycle matrices etc. — see
+  # scrape/vision.py). Uses the host's primary Ollama on :11434, the only one
+  # with a vision model; the embed pool above is nomic-only. qwen2.5vl:7b
+  # scored best on the release-schedule matrix (55-56/56, ~18s, GPU-resident,
+  # prompt-robust). Content-hash cached in corpus/.vision-cache/; VISION_MAX_NEW
+  # bounds NEW OCRs per run so the cache fills incrementally instead of a
+  # multi-hour first pass. Any failure degrades to the pre-vision behavior.
+  VISION_OCR: "1"
+  VISION_URL: http://192.168.0.2:11434
+  VISION_MODEL: qwen2.5vl:7b
+  VISION_MAX_NEW: "60"
+
   PRODUCT_NAME: morpheus
 
 jobs:
@@ -64,6 +76,9 @@ jobs:
         run: |
           python -m pip install -q --upgrade pip
           python -m pip install -q -r requirements.txt
+          # Vision OCR deps (Pillow) — only the scrape step needs these; kept
+          # out of requirements.txt so they never bloat the server image.
+          python -m pip install -q -r requirements-vision.txt
 
       # ---- Phase 1: scrape ---------------------------------------
       - name: Refresh bundle catalog