AP_Math: remove warnings from constrain_value()

Return type is T which can be an integral type, float or double. By
dividing by 2 we avoid float operation on the first case and do the
right thing on the second and third.

libraries/AP_Math/AP_Math.cpp

@@ -176,10 +176,10 @@ template <class T>
 T constrain_value(const T amt, const T low, const T high)
 {
     // the check for NaN as a float prevents propagation of floating point
-    // errors through any function that uses constrain_float(). The normal
+    // errors through any function that uses constrain_value(). The normal
     // float semantics already handle -Inf and +Inf
     if (isnan(amt)) {
-        return (low + high) * 0.5f;
+        return (low + high) / 2;
     }
 
     if (amt < low) {