Math: workaround MSVC2015's inability to pick up the float*Unit operator.

Basically the same workaround as was done for Vector in
9cb623eaf9, except that here it doesn't
need the macros for subclasses.

src/Magnum/Math/Unit.h

@@ -184,6 +184,10 @@ template<template<class> class Derived, class T> class Unit {
          * Similar to @ref operator*(T) const, except that the multiplication
          * is done in floating-point.
         */
+        /* Note that this one isn't correctly picked up on MSVC 2015, there's
+           an out-of-class overload wrapped in CORRADE_MSVC2015_COMPATIBILITY
+           which is (and the two don't conflict, apparently, so both are
+           present) */
         #ifdef DOXYGEN_GENERATING_OUTPUT
         template<class FloatingPoint> constexpr Unit<Derived, T>
         #else
@@ -292,6 +296,18 @@ template<template<class> class Derived, class T> class Unit {
         T _value;
 };
 
+#ifdef CORRADE_MSVC2015_COMPATIBILITY
+/* MSVC 2015 doesn't correctly pick up the in-class inline friend that does
+   this, resulting in float*Unit<int> expressions being wrongly executed as
+   int*Unit<int> due to an implicit conversion fallback. This overload is
+   picked up correctly (and doesn't conflict with the in-class one). See
+   UnitTest::multiplyDivideIntegral() for regression tests, the same issue and
+   a matching workaround is done in Vector as well. */
+template<template<class> class Derived, class FloatingPoint, class Integral> constexpr typename std::enable_if<std::is_integral<Integral>::value && std::is_floating_point<FloatingPoint>::value, Unit<Derived, Integral>>::type operator*(FloatingPoint number, const Unit<Derived, Integral>& value) {
+    return value*number;
+}
+#endif
+
 }}
 
 #endif

src/Magnum/Math/Vector.h

@@ -1509,7 +1509,8 @@ extern template MAGNUM_EXPORT Debug& operator<<(Debug&, const Vector<4, Double>&
    need to use the MAGNUM_VECTORn_OPERATOR_IMPLEMENTATION() overloads as well
    to return a correct subtype. See VectorTest::multiplyDivideIntegral(),
    VectorTest::subclass() and corresponding cases in Vector2Test, Vector3Test,
-   Vector4Test and ColorTest for regression tests. */
+   Vector4Test and ColorTest for regression tests. The same issue and a
+   matching workaround is done in Unit as well. */
 template<std::size_t size, class FloatingPoint, class Integral> constexpr typename std::enable_if<std::is_integral<Integral>::value && std::is_floating_point<FloatingPoint>::value, Vector<size, Integral>>::type operator*(FloatingPoint scalar, const Vector<size, Integral>& vector) {
     return vector*scalar;
 }