Previously check() re-fetched the configured image URL over HTTP on
every poll (every 30 min by default). If the remote image ever came
back even slightly byte-different between fetches -- common with
dynamically-generated images -- that forced a real e-ink flash for no
visible reason, since this panel has no partial refresh.
Now the URL is downloaded exactly once, server-side in the settings
route, and saved into web/static/uploads/current.<ext> the same way a
direct upload already was -- clearing any previous upload/download
first so switching sources doesn't accumulate orphaned files.
check() just reads that local file regardless of whether it came from
an upload or a URL, with zero network calls. Also added a browser
User-Agent to the download request -- confirmed live that Wikimedia
(plausibly a common source people would link to) 403s the default
requests/python User-Agent as bot traffic, same class of issue as the
EFA transit endpoints handled earlier.
Verified end-to-end against real URLs: download+cache works, a second
URL replaces rather than accumulates, and a settings-only save (no
new URL entered) no longer wipes the cached picture.
Honey
5889422c31