fix(images): percent-encode original-image URLs ('#' in paths 404'd)

An image whose on-disk path contains '#' (post folders like 'BLUE#59')
served its hash-named thumbnail fine but 404'd the original: the unencoded
'#' in image_url was parsed by the browser as a URL fragment, so
'#59/01_timelapse.jpg' never reached the /images route. Add a shared
image_url(path) helper that percent-encodes the path (safe='/') and route
the 3 raw builders (gallery detail + 2 in series) through it. Not a
cleanup-tool deletion — the file is on disk; only the URL was wrong.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tests/test_gallery_service.py

@@ -8,11 +8,22 @@ from backend.app.services.gallery_service import (
     GalleryService,
     decode_cursor,
     encode_cursor,
+    image_url,
 )
 
 pytestmark = pytest.mark.integration
 
 
+def test_image_url_percent_encodes_special_chars():
+    # A '#' in a post folder ('BLUE#59') would otherwise be parsed as a URL
+    # fragment, dropping the rest of the path and 404'ing the original while the
+    # hash-named thumbnail still loads. Operator-flagged 2026-06-12.
+    url = image_url("/images/dismassd/patreon/2024-04-28_BLUE#59/01 a.jpg")
+    assert url == "/images/dismassd/patreon/2024-04-28_BLUE%2359/01%20a.jpg"
+    # '/' stays a separator; a plain path is unchanged.
+    assert image_url("/images/a/b/c.jpg") == "/images/a/b/c.jpg"
+
+
 def _now():
     return datetime.now(UTC)