Also no longer disable run-time CPU detection for fpu targets except
arm which is built for each variant (neon, vfp).
This vastly improves performance on x86 which now supports building ASM
code with yasm and makes use of a broad spectrum of extensions
depending on their availability at run-time.
In future, such differentiation should also be introduced for MIPS
to at least conditionally enable the output of MIPS32r2 instructions as
well as DSPr1 and DSPr2 if the target supports that dispite them having
CONFIG_SOFT_FLOAT=y set.
Signed-off-by: Daniel Golle <daniel@makrotopia.org>