Arm64: leave .bfloat16 processing to common code

With x86 support having been implemented by extending atof-ieee.c, avoid
unnecessary code duplication in md_atof(). This will then also allow to
take advantage of adjustments made there without needing to mirror them
here.