Math: updated fuzzy comparison epsilon values for long double and double.

To be consistent with what's for floats. Maybe I'll need to decrease the
precision later again, but at least this has now some real background
instead of being a guesstimate.

src/Magnum/Math/TypeTraits.h

@@ -33,22 +33,34 @@
 
 #include "Magnum/Types.h"
 
-/** @brief Precision when testing floats for equality */
+/**
+@brief Precision when testing floats for equality
+
+They have "at least" 6 significant digits of precision, taking one digit less
+for more headroom.
+*/
 #ifndef FLOAT_EQUALITY_PRECISION
 #define FLOAT_EQUALITY_PRECISION 1.0e-5f
 #endif
 
-/** @brief Precision when testing doubles for equality */
+/**
+@brief Precision when testing doubles for equality
+
+They have "at least" 15 significant digits of precision, taking one digit less
+for more headroom.
+*/
 #ifndef DOUBLE_EQUALITY_PRECISION
-#define DOUBLE_EQUALITY_PRECISION 1.0e-12
+#define DOUBLE_EQUALITY_PRECISION 1.0e-14
 #endif
 
 /**
 @brief Precision when testing long doubles for equality
-@todo some proper value please
+
+They have "at least" 18 significant digits of precision, taking one digit less
+for more headroom.
 */
 #ifndef LONG_DOUBLE_EQUALITY_PRECISION
-#define LONG_DOUBLE_EQUALITY_PRECISION 1.0e-18l
+#define LONG_DOUBLE_EQUALITY_PRECISION 1.0e-17l
 #endif
 
 namespace Magnum { namespace Math {