Remove truncating version of kai_cast_bf16_f32

* The implementation of kai_cast_bf16_f32 in kai_common.h
  violates the one definition rule (ODR). Depending on the
  archirectural feature enabled by the compiler, the function
  is implemented differently.
* In the kernel library, kai_cast_bf16_f32 is only used by
  microkernels that are compiled with BF16 enabled, therefore
  this change doesn't affect the functionality of the microkernel.
* The example code needs to be compiled with BF16 enabled to match
  with the behavior of the microkernel.

Signed-off-by: Viet-Hoa Do <viet-hoa.do@arm.com>