loop: shared bounded-turn core accepts history + knobs (ServeTurn)

Turn machinery refactored so the upcoming serve front door reuses it
instead of copying it: the round loop is now runTurn over arbitrary
message history with an optional tool palette and forwarded max_tokens/
temperature. The serve path (ServeTurn) is stateless chat over client
history with tools hard-off: a tool call the model produces anyway is
refused as a tool result, never executed. Also exposes the model router
(so serve can build its catalog) and a sorted class list; llm requests
can now carry temperature.

💘 Generated with Crush

Assisted-by: Crush:glm-5.2

internal/models/models.go

@@ -66,3 +66,14 @@ func (r *Router) Tiers() []string {
 	sort.Strings(out)
 	return out
 }
+
+// Classes returns the sorted task classes known to the router (the
+// `harness serve` model catalog is derived from it: one model per class).
+func (r *Router) Classes() []string {
+	out := make([]string, 0, len(r.classes))
+	for k := range r.classes {
+		out = append(out, k)
+	}
+	sort.Strings(out)
+	return out
+}

internal/models/models_test.go

@@ -76,3 +76,13 @@ func TestNewRouterValidation(t *testing.T) {
 		t.Errorf("empty tiers must error")
 	}
 }
+
+func TestClasses(t *testing.T) {
+	// The serve front door builds its catalog from the sorted class list;
+	// it must be complete and ordered for stable /v1/models output.
+	got := testRouter(t).Classes()
+	want := "code primary read review study"
+	if strings.Join(got, " ") != want {
+		t.Errorf("Classes() = %v, want %v", got, want)
+	}
+}