Fix intrinsic collisions due to NEON flag usage

[stellar-aria]

See #220 for details.

Essentially there's function/macro definition collisions that occur due to already being defined in cmsis_gcc.h.

I'm not entirely sure if this is the right way to fix this issue, since there's a note about CMSIS/Core not being available on Cortex-A/R, ...except that it seems to be the only way to build right now on a baremetal/non-hosted system, since otherwise there's complaints about cmsis_compiler.h missing in the same file.

[christophe0606]

The problem is that some old cores (and I think some Cortex-R) have Neon and DSP extensions but they are not supported by CMSIS-Core.

The original and this PR will both fail to work on some core.

So I think __GNUC_PYTHON__ should be used in the test (it is in fact a flag to check if CMSIS Core must be included or not).

When __GNUC_PYTHON__ is enabled I should never define ARM_MATH_DSP to ensure no intrinsics is used if the CMSIS Core is not available even if __ARM_FEATURE_DSP is defined.