[aarch64] Fix fmin and fmax on aarch64 (#140)

vmaxnum_v*_v*_v* and vminnum_v*_v*_v* are helper functions for utilizing instructions for choosing larger and smaller elements from two given vectors, and they are introduced in the following PR.

https://github.com/shibatch/sleef/pull/109

It is said that vmaxnmq and vminnmq on aarch64 are IEEE754-conformant, but I recently found that handling of signalling NaN by these instructions is not conforming to the specification of fmin and fmax functions in the ANSI C standard.

This patch fixes that problem. It also adds regression test for checking signaling nan handling in fmin and fmax.