psq_l with QUANTIZE_FLOAT does not use the FPU, so it does not trim the precision of the u32 input data. We already have the helper ConvertToDouble for floating point u32->u64 convertion used in lfs, so let's use it here as well.