fix(retrieval): the newest two rows are not the newest row of each dial (#4104)

`current_settings` read a surface's history with one query and `limit(2)`, then
keyed the rows by dial. That is only the same thing while both dials have moved
equally often — and they do not. Floors get walked; budgets rarely move. Three
floor changes and one budget change returns two floor rows, and the budget
change vanishes.

Under #4102 that cost a missing reason. Since #4104 it is worse: the dial then
reports `source: "shipped"` — still on the value Scribe shipped — for a number
somebody deliberately tuned. A wrong calibration answer, in the direction a
reader has no cause to double-check.

One query per dial, `limit(1)` each.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01821k5B3Ysecp9fNYs92Kuy

tests/test_calibration_stamp.py

@@ -189,3 +189,50 @@ def test_the_tool_says_nothing_auto_retunes():
     assert "calibration" in doc
     assert "NOTHING IS RETUNED AUTOMATICALLY" in doc
     assert "unstamped" in doc
+
+
+@pytest.mark.asyncio
+async def test_a_dial_is_read_per_dial_not_from_the_newest_two_rows():
+    """The floor's history must not be able to bury the budget's.
+
+    "The newest two rows" and "the newest row of each dial" differ the moment
+    one dial moves more often than the other — which is the normal case, since
+    floors get walked and budgets rarely do. Before this was one query per dial,
+    a surface with three floor changes and one budget change reported the budget
+    as untouched: a WRONG calibration answer rather than a missing one, and
+    wrong in the direction a reader would not think to check.
+    """
+    session = make_mock_session()
+    per_dial = {
+        "floor": MagicMock(dial="floor", embedding_model="old/model",
+                           shape_version=1,
+                           to_dict=MagicMock(return_value={"dial": "floor"})),
+        "budget": MagicMock(dial="budget", embedding_model="old/model",
+                            shape_version=1,
+                            to_dict=MagicMock(return_value={"dial": "budget"})),
+    }
+    calls = []
+
+    def execute(stmt):
+        # The dial is whichever the WHERE clause names; a query that did not
+        # scope by dial would render this stand-in unable to answer, which is
+        # the point.
+        sql = str(stmt.compile(compile_kwargs={"literal_binds": True}))
+        dial = "budget" if "'budget'" in sql else "floor"
+        calls.append(dial)
+        result = MagicMock()
+        result.scalars.return_value.first.return_value = per_dial[dial]
+        return result
+
+    session.execute = AsyncMock(side_effect=execute)
+    with patch.object(rt, "async_session", MagicMock(return_value=session)), \
+         patch.object(rt, "floor_for", AsyncMock(return_value=0.7)), \
+         patch.object(rt, "budget_for", AsyncMock(return_value=3)):
+        out = await rt.current_settings(1)
+
+    assert calls.count("floor") == len(SURFACES)
+    assert calls.count("budget") == len(SURFACES)
+    for row in out:
+        # Both dials tuned under a model that is not the live one.
+        assert row["calibration"]["budget"]["source"] == "tuned"
+        assert row["calibration"]["budget"]["stale"] is True