Relax tolerance on AmericanOption test

The test tolerance is very strict and even small changes in floating point
order causes the test to fail (actual error is ~5e-15 rather than required 1e-15).
It is highly unlikely that failure in this case indicates invalid assembly.
This patch allows both AOCC with -O3 -zopt -amdveclib and ICPX with -O3 -xCORE-AVX512
-fp-model=precise to pass.

test-suite/americanoption.cpp

@@ -1542,7 +1542,7 @@ BOOST_AUTO_TEST_CASE(testQdEngineStandardExample) {
         );
         const Real calculated = americanOption.NPV() - europeanOption.NPV();
 
-        const Real tol = 1e-15;
+        const Real tol = 7e-15;
         const Real diff = std::abs(calculated - expected[i]);
 
         if (diff > tol) {