Audio feature math (shared core)#

Shared sl_fe_audio_config_t and toolchain-neutral, side-effect-free feature-extraction math used by both platform implementations. Quantizer and normalization helpers must stay bit-identical across targets, so they live in a single plain-C translation unit.

Each feature accessor reads count elements from a circular feature buffer in logical order: element i is feature_buffer[(start + i) % count]. The caller owns validation (NULL checks, element-count checks, parameter ranges) and ring-index bookkeeping; these helpers only compute outputs.