[Bug target/122014] (AArch64) Optimize 8-bit and 16-bit popcount as special cases

[Explorer09 at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122014

--- Comment #3 from Kang-Che Sung <Explorer09 at gmail dot com> ---
Acknowledge the popcount_8_bp and popcount_16_bp case.

The point is to recognize the don't-care so that gcc won't need to mask the
upper bits in order to constrain the values to `uint8_t` or `uint16_t`.

Also, because the ADDV instruction adds the values of *all* lanes in a vector,
for the 16-bit case, it would be better to use UADDLP instead as UADDLP would
effectively make the upper bit ignored after extracting the sum of lane 0.