refs: Flickr + web search for the hero entrance

common_fetch was out of material, so an OpenRouter web-search call (gpt-5.2,
openrouter:web_search) returned candidate photo pages and page_image_fetch.py
resolved each og:image and downloaded it, recording page URL and hash locally.
Four frames kept, including the clearest entrance view yet.

The frames overturn the inherited 'eight stone steps, 12 m wide': the entrance
is three portals with individual short flights. Recorded as a resolved conflict
with the photographs as evidence.

Tools/README.md

@@ -17,6 +17,7 @@ Repeatable import/export and processing helpers. Anything that was done twice by
 - `street_mesh.py` — road, kerbs, and sidewalks on the DEM, in the project frame; kerbs flush at the three crossings.
 - `terrain_from_dem.py` — terrain grid from the committed DEM, with building footprints cut out.
 - `blosm_blockout.py` — headless Blosm import of the committed `.osm` extract into a `.blend`, with a per-building scale report. Exact invocation in `Art/blender/README.md`.
+- `page_image_fetch.py` — resolves a web page's `og:image` and downloads it, with page URL and hash in a local sidecar. Works for Flickr, Commons file pages, and most photo blogs.
 - `commons_fetch.py` — Wikimedia Commons reference fetcher. Search, exact `File:` titles, or a whole Commons category (`--category`), thumbnails, provenance `sidecar.json` per folder. Wikimedia throttles `upload.wikimedia.org` (HTTP 429), so keep batches to six or fewer files and pass `--delay 3`; a throttled file is skipped and recorded rather than aborting the run. Writes only into the gitignored `Art/reference/`; reference images are never committed.
 
 ## Planned helpers