CMSIS-NN: Cleanup of CMSIS-DSP dependencies

Left over CMSIS-DSP dependencies are removed

For functions with just DSP extension optimization,
the reference C implementation(if available) is picked
for processors with Helium Technology as auto vectorization
can potentially give a better performance. This affects only
the non-TFLM functions.

Change-Id: I8ededad1d34eeb27c1ac2d65b3250b85562cc0d4