Adding a site's address subscribes to the feed it links (#71, #72)

Both add paths, the CLI's and the web's, look behind the URL first: a web
page that names its feed with <link rel="alternate"> is swapped for that
feed, before the duplicate check so it finds a feed someone already has.
Before, the page itself was added and every scan failed on it.

alternate_feed_link found tags in a to_lowercase() copy and sliced the
original at those offsets; Unicode lowercasing changes some characters'
length, so a page with one before its <link> tags lost the href or
panicked off a char boundary. ASCII lowercasing keeps offsets aligned.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

src/web.rs

@@ -1279,7 +1279,8 @@ async fn add_feed(
     Json(body): Json<NewFeed>,
 ) -> Result<Json<serde_json::Value>, ApiError> {
     let mut cfg = (*state.ctx.cfg()).clone();
-    let url = crate::feed::expand_input(&body.url);
+    // Before the duplicate check, so a site's page finds the feed someone already has.
+    let url = crate::feed::feed_behind_page(&state.ctx.client, &crate::feed::expand_input(&body.url)).await;
     // Someone else may already have it. Then adding costs nothing: no second fetch, no
     // second copy on disk, just another name against the same feed.
     if let Some(existing) = crate::subscriptions(&state.ctx).await?