For float4/8, remove errno checks for pow() and exp() because only some

platforms set errno, and we already have a check macro that detects
under/overflow, so there is no reason for platform-specific code
anymore.