Specify FMax/FMin float intrinsics for riscv64

Removes the generalized "common" implementations for FMax and FMin and
defines them specifically for riscv64 according to IEEE 754-2019.

Updates vector interpreter tests to check this behavior explicitly.

Bug: 330396676
Test: Bionic unit tests: math_h.{fmax,fmin,fmaxf,fminf}
Test: berberis_host_tests
Change-Id: Ia82e441432e22512686b632bfc4a0179d521e46a
3 files changed