Implement `BMat16::nr_rows`

[james-d-mitchell]

inline size_t BMat16::nr_rows() const noexcept {
  [...] 
  //// Vectorized version which doesn't work due to the absence of popcnt in
   /// simde
   // xpu16 tmp = _data, zero = simde_mm256_setzero_si256();
   // xpu16 x = (tmp != zero);
   // return simde_mm256_popcnt_epi16(x);
}

As soon as popcnt is implemented in simde