Make an unit test of std.math IEEE 754-2008 compliant.

IEEE 754-2008 does not specify the sign bit of a NaN. The exceptions to this
rule are the operations copy, negate, abs and copySign. In all other cases,
the standard does not specify the sign bit (chapter 6.3 on page 35).

The unit test I like to change with this pull request assumes that the sign
bit of a NaN is propagated through the exp() computation. This assumption
violates IEEE 754-2008.

My fix is the use of abs() on the result, because this operation changes the
sign bit of a NaN in a compliant way.

There might be other places which must be fixed but this one really hurts me.

Author: redstar

std/math.d

@@ -1636,13 +1636,13 @@ unittest
     resetIeeeFlags();
     x = exp(real.nan);
     f = ieeeFlags;
-    assert(isIdentical(x,real.nan));
+    assert(isIdentical(abs(x), real.nan));
     assert(f.flags == 0);
 
     resetIeeeFlags();
     x = exp(-real.nan);
     f = ieeeFlags;
-    assert(isIdentical(x, -real.nan));
+    assert(isIdentical(abs(x), real.nan));
     assert(f.flags == 0);
 
     x = exp(NaN(0x123));