Unify Geometry_SSE.h into a generic version and make it support both for x86 and ARM

The quat_product and quat_conj ops in Geometry_SSE.h just supported for x86,i am working on replacing it by a generic version and making it supported both for x86 and arm.We also benchmarked this and confirmed that the generic version does not bring any regression on x86, but get a great improvement on ARM.Code for testing and benchmarks belows:

#include<benchmark/benchmark.h>
#include<Eigen/Geometry>

using namespace Eigen;

template <typename T> inline void DoNotOptimizeAway(T&& datum) {
    asm volatile ("" : "+m,r" (datum): :"memory");
}

static void quat_product_f(benchmark::State& state){
  Quaternion<float,AutoAlign> q1,q2,q3;
  q1.coeffs().setRandom();
  q2.coeffs().setRandom();
  for(auto _ : state){
    DoNotOptimizeAway(q3 = q2 * q1);
  }
}

static void quat_conj_f(benchmark::State& state){
  Quaternion<float,AutoAlign> q1,q2;
  q1.coeffs().setRandom(); 
  q2.coeffs().setRandom(); 
  for(auto _ : state){
    DoNotOptimizeAway(q2 = q1.conjugate());    
  }
}

static void quat_product_d(benchmark::State& state){
  Quaternion<double,AutoAlign> q1,q2,q3;
  q1.coeffs().setRandom();
  q2.coeffs().setRandom();
  for(auto _ : state){
    DoNotOptimizeAway(q3 = q2 * q1);
  }
}

static void quat_conj_d(benchmark::State& state){
  Quaternion<double,AutoAlign> q1,q2;
  q1.coeffs().setRandom(); 
  q2.coeffs().setRandom(); 
  for(auto _ : state){
    DoNotOptimizeAway(q2 = q1.conjugate());    
  }
}


BENCHMARK(quat_product_f);
BENCHMARK(quat_conj_f);

BENCHMARK(quat_product_d);
BENCHMARK(quat_conj_d);

BENCHMARK_MAIN();

We used GCC to compile and turn on the -O3 optimization option.
Benchmarks belows:

compare the generic version to SSE version on x86
----------------------------------------------
Benchmark           OldTime         NewTime   
----------------------------------------------
quat_product_f       2.21 ns         2.21 ns   
quat_conj_f         0.418 ns        0.428 ns   
quat_product_d       2.84 ns         2.84 ns   
quat_conj_d         0.631 ns        0.632 ns   

compare the generic version to STD version(Not Vectorized) on ARM
-----------------------------------------------
Benchmark            OldTime         NewTime  
-----------------------------------------------
quat_product_f       9.30 ns         5.08 ns   
quat_conj_f          8.49 ns         1.13 ns    
quat_product_d       9.60 ns         9.62 ns    
quat_conj_d          8.49 ns         2.05 ns