Add personal notes, DeepSearch history, backup offload, and gateway auth throttling

Also streamline the top navigation: drop the Tools dropdown, make Quizzes
and Battles icon-only entries, and remove the Workspace, Containers and
Editor entry points from the project detail page.

tests/unit/services/quiz/grading.py

@@ -134,8 +134,9 @@ def test_grade_free_text_uses_a_valid_verdict(monkeypatch):
 def test_grade_free_text_passes_the_answering_key_through(monkeypatch):
     seen = {}
 
-    def capture(api_key, system, text, timeout, model=None):
+    def capture(api_key, system, text, timeout, model=None, bypass_preamble=False):
         seen["api_key"] = api_key
+        seen["bypass_preamble"] = bypass_preamble
         return json.dumps({"score": 1.0}), None
 
     monkeypatch.setattr(grading, "gateway_complete", capture)
@@ -143,6 +144,18 @@ def test_grade_free_text_passes_the_answering_key_through(monkeypatch):
     assert seen["api_key"] == "member-key"
 
 
+def test_grade_free_text_bypasses_the_operator_preamble(monkeypatch):
+    seen = {}
+
+    def capture(api_key, system, text, timeout, model=None, bypass_preamble=False):
+        seen["bypass_preamble"] = bypass_preamble
+        return json.dumps({"score": 1.0}), None
+
+    monkeypatch.setattr(grading, "gateway_complete", capture)
+    grading.grade_free_text("member-key", QUESTION, "answer")
+    assert seen["bypass_preamble"] is True
+
+
 def test_the_fallback_score_is_the_deterministic_overlap():
     expected = scoring.token_overlap_score(QUESTION["expected_answer"], "one bucket index")
     result = grading.grade_free_text("", QUESTION, "one bucket index")