arm/translate-a64: implement half-precision F(MIN|MAX)(V|NMV)

arm/translate-a64: implement half-precision F(MIN|MAX)(V|NMV)

This implements the half-precision variants of the across vector
reduction operations. This involves a re-factor of the reduction code
which more closely matches the ARM ARM order (and handles 8 element
reductions).

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Message-id: 20180227143852[email protected]
Signed-off-by: Peter Maydell <[email protected]>