test(régression): le fake reproduit le payload Gitea réel (pull_request: null)

Le faux omettait la clé pull_request sur les issues, ce qui a
masqué le bug de la 0.5.1 pendant que 84 tests restaient verts.
Clé désormais toujours présente (null = issue, objet = PR) +
4 tests dédiés dans test_regression_gitea_payload.py.

tests/conftest.py

@@ -55,9 +55,10 @@ def _issue(number: int, title: str, state: str = "open", pr: bool = False) -> di
         "created_at": "2026-09-01T09:00:00Z",
         "updated_at": "2026-09-02T09:00:00Z",
         "html_url": f"http://gitea.test/Projets/demo/issues/{number}",
+        # Faithful to real Gitea 1.27 payloads: the key is ALWAYS
+        # present — null for a plain issue, an object for a PR.
+        "pull_request": {"url": "http://gitea.test/pr"} if pr else None,
     }
-    if pr:
-        issue["pull_request"] = {"url": "http://gitea.test/pr"}
     return issue