news: harden paywall exclusion at the candidate query + add the missing regressions

Codex's two non-blocking hardening items, folded in before cutover:
- _candidate_articles() now excludes paywalled sources IN-QUERY (before LIMIT 50),
  so flagged stories can't consume candidate slots and leave a full brief thin.
  Dropped the now-redundant post-fetch filter in build_daily_brief.
- Regressions: history retains a viewed paywalled article; sitemap omits a
  paywalled source AND restores it under override="free".
- Aligned test_brief_paywall to the source-level model (paywalled sources carry a
  paywalled homepage, as in production) — it had relied on article-URL detection.

425 backend tests green.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

tests/test_brief_paywall.py

@@ -8,9 +8,12 @@ from goodnews.paywall import is_paywalled
 def test_brief_prefers_readable_over_higher_scored_paywalled():
     c = connect(":memory:"); init_db(c)
     today = date.today().isoformat()
+    # Sources 1-2 are paywalled by their homepage domain (matches production: a paywalled
+    # source has a paywalled site); 3-7 are free.
+    homes = {1: "https://www.newscientist.com/", 2: "https://www.nature.com/"}
     for sid in range(1, 8):
-        c.execute("INSERT INTO sources (id,name,feed_url,trust_score) VALUES (?,?,?,5)",
-                  (sid, f"S{sid}", f"http://s{sid}/f"))
+        c.execute("INSERT INTO sources (id,name,feed_url,homepage_url,trust_score) VALUES (?,?,?,?,5)",
+                  (sid, f"S{sid}", f"http://s{sid}/f", homes.get(sid)))
 
     def add(aid, sid, url, score):
         c.execute("INSERT INTO articles (id,source_id,canonical_url,title,published_at,url_hash) "
@@ -19,7 +22,7 @@ def test_brief_prefers_readable_over_higher_scored_paywalled():
                   "cortisol_score,ragebait_score,pr_risk_score,accepted,topic,flavor) "
                   "VALUES (?,?,?,?,0,0,2,1,'science','discovery')", (aid, score, score, score))
 
-    # Paywalled ones are scored HIGHER — readability must still win for the five.
+    # Paywalled sources are scored HIGHER — they must still be EXCLUDED, leaving the five readable.
     add(1, 1, "https://www.newscientist.com/a", 9)
     add(2, 2, "https://www.nature.com/b", 9)
     add(3, 3, "https://phys.org/c", 4)
@@ -33,4 +36,4 @@ def test_brief_prefers_readable_over_higher_scored_paywalled():
     urls = [r["canonical_url"] for r in show_brief(c, brief_date=today, limit=10)]
     c.close()
     assert len(urls) == 5
-    assert not any(is_paywalled(u) for u in urls)  # five readable chosen over paywalled
+    assert not any(is_paywalled(u) for u in urls)  # paywalled sources excluded; five readable fill