We return uint32_t instead of uint16_t because that seems to be more efficient for most purposes (cutting it down to uint16_t costs performance in som

Convert num's each digit as packed 16-bit in a vector. num's digits as abcdefgh (high bits is 0 if not enough) The converted vector is { a, b, c, d, e

Convert num {abcd} to {axxxx, abxxxx, abcxxxx, abcdxxxx}