Float to int32 conversion may cause overflow

[jdemel]

There's an issue in volk_32f_s32f_convert_32i.

The intended logic is:

1. scale the input float value by scalar
2. clamp float value to 32bit integer range [-2147483648, 2147483647], or [-2147483648, 2147483648)
3. Convert to int.

Unfortunately, the operation:

float value = 2147483648.0f - 1.0f;

does not result in 2147483647.0f, but 2147483648.0f since the next lower float representable number is 2147483520.0f. Thus,

int32_t result = (int32_t) 2147483648.0f - 1.0f;

results in -2147483648, aka overflow.

It may be possible to clamp the float value to 2147483520.0f but a number such as 6167483520.0f would then result in 2147483520 instead of 2147483647 which would've been the expected value.