refactor: Reimplement _mm_movelh_ps for Arm64

[howjmay]

Accelerate _mm_movelh_ps for about 6.71%.
Test benchmark is listed below.

old average time: 4.062780e-05
new average time: 3.790102e-05
old time std: 1.375459e-06
new time std: 1.348056e-06

The benchmark was generated by following test

    const float *_a = impl.mTestFloatPointer1;
    const float *_b = impl.mTestFloatPointer2;
    __m128 a = load_m128(_a);
    __m128 b = load_m128(_b);

    clock_t t;
    double time_taken;
    t = clock();
    for (int i = 0; i < 10000; i++) {
        old_mm_movehl_ps(a, b);
    }
    t = clock() - t;
    time_taken = ((double) t) / CLOCKS_PER_SEC;
    printf("old: %f\n", time_taken);

    t = clock();
    for (int i = 0; i < 10000; i++) {
        new_mm_movehl_ps(a, b);
    }
    t = clock() - t;
    time_taken = ((double) t) / CLOCKS_PER_SEC;
    printf("new: %f\n", time_taken);