Fix a negative-zero issue in complex numbers; make all portable tests pass

src/Numerics/Complex32.cs

@@ -169,8 +169,9 @@ namespace MathNet.Numerics
         /// <returns>The phase or argument of this <c>Complex32</c></returns>
         public float Phase
         {
+            // NOTE: the special case for negative real numbers fixes negative-zero value behavior. Do not remove.
             [TargetedPatchingOptOut("Performance critical to inline this type of method across NGen image boundaries")]
-            get { return (float)Math.Atan2(_imag, _real); }
+            get { return _imag == 0f && _real < 0f ? (float)Constants.Pi : (float)Math.Atan2(_imag, _real); }
         }
 
         /// <summary>

src/Numerics/Complex64.cs

@@ -162,8 +162,9 @@ namespace MathNet.Numerics
         /// <returns>The phase or argument of this <c>Complex</c></returns>
         public double Phase
         {
+            // NOTE: the special case for negative real numbers fixes negative-zero value behavior. Do not remove.
             [TargetedPatchingOptOut("Performance critical to inline this type of method across NGen image boundaries")]
-            get { return Math.Atan2(_imag, _real); }
+            get { return _imag == 0d && _real < 0d ? Constants.Pi : Math.Atan2(_imag, _real); }
         }
 
         /// <summary>

src/UnitTests/LinearAlgebraTests/Complex/Factorization/QRTests.cs

@@ -707,7 +707,7 @@ namespace MathNet.Numerics.UnitTests.LinearAlgebraTests.Complex.Factorization
 
             for (var i = 0; i < vectorX.Count; i++)
             {
-                AssertHelpers.AlmostEqual(test[i], vectorX[i], 9);
+                AssertHelpers.AlmostEqualAbsolute(test[i], vectorX[i], 9);
             }
 
             // Make sure A didn't change.