Correct Q31 scalar output saturation documentation

[sylvesterkaczmarek]

Addresses #328

Summary

• correct the final-conversion documentation for nine Q31 filtering and matrix functions
• state that the scalar paths convert their 2.62 accumulators without saturation
• describe the LMS and normalized LMS shift in terms of postShift
• leave arm_mat_cmplx_mult_q31 unchanged because its scalar path calls clip_q63_to_q31

Context

The affected comments currently promise saturation while the scalar implementations store a shifted accumulator through a plain q31_t conversion. Issue #175 confirms that non-saturating Q31 behavior is intentional for scalar kernels and also notes that Helium can saturate. The revised wording is therefore explicitly scoped to the scalar implementations.

Validation

• scripted source audit across all nine files
• confirmed every affected comment now states the non-saturating scalar behavior
• confirmed none of the nine files retain the reported saturation claims
• confirmed the saturating arm_mat_cmplx_mult_q31 exception remains unchanged
• git diff --check

The full CMSIS-DSP test suite was not run because this change only updates Doxygen comments and does not alter compiled code.