patching MacOS performance

[TysonRayJones]

as per #638

[TysonRayJones]

The below quick-and-dirty script did not reveal an accuracy difference when using -Ofast in Clang:

int main() {
    initQuESTEnv();
    setMaxNumReportedSigFigs(20);

    // start in |+>
    Qureg qureg = createQureg(20);
    initPlusState(qureg);

    qreal angle = 0.3;

    // repeatedly effect R(a)R(-a) = I
    for (int q=0; q<qureg.numQubits; q++) {
        applyRotateX(qureg, q,   angle);
        applyRotateX(qureg, q, - angle);
    }

    // we should return to |+> with prob=0.5
    qreal prob = calcProbOfQubitOutcome(qureg, 0, 0);
    reportScalar("prob", prob);

    destroyQureg(qureg);
    finalizeQuESTEnv();
    return 0;
}


// -Ofast: prob = 0.49999999999999994449
// else:   prob = 0.49999999999999994449 (unchanged!)