The optimized instruction sequences for floating-point min/max do not
have the same behaviour as TurboFan's Float(32|64)(Min|Max) functions
(incorrect handling for NaN operands).

R=dstence@us.ibm.com, michael_dawson@ca.ibm.com
BUG=

Review URL: https://codereview.chromium.org/1193843015

Cr-Commit-Position: refs/heads/master@{#29234}