simd, sse: use 0x80000000 insteaf of -0.f to fix fastmath on msvc

[recp]

fixes #289

Now we have global macro for sign masks:

#define GLMM_NEGZEROf 0x80000000 /*  0x80000000 ---> -0.0f  */

#define GLMM__SIGNMASKf(X, Y, Z, W)                                           \
   _mm_castsi128_ps(_mm_set_epi32(X, Y, Z, W))
  /* _mm_set_ps(X, Y, Z, W); */

#define glmm_float32x4_SIGNMASK_PNPN GLMM__SIGNMASKf(0, GLMM_NEGZEROf, 0, GLMM_NEGZEROf)
#define glmm_float32x4_SIGNMASK_NPNP GLMM__SIGNMASKf(GLMM_NEGZEROf, 0, GLMM_NEGZEROf, 0)
#define glmm_float32x4_SIGNMASK_NPPN GLMM__SIGNMASKf(GLMM_NEGZEROf, 0, 0, GLMM_NEGZEROf)

#define glmm_float32x4_SIGNMASK_NEG _mm_set1_ps(GLMM_NEGZEROf)

for instance:
_mm_set_ps(-0.f, 0.f, -0.f, 0.f) -> glmm_float32x4_SIGNMASK_NPNP
_mm_set_ps(0.f, -0.f, 0.f, -0.f) -> glmm_float32x4_SIGNMASK_PNPN

[recp]

@myfreeer we also need wasm update to use these macros ( or better names can be used if we get enough feedbacks ):

#define glmm_float32x4_SIGNMASK_PNPN ...
#define glmm_float32x4_SIGNMASK_NPNP ...
#define glmm_float32x4_SIGNMASK_NPPN ...
 ...

I can leave this update to you if you'd like :)