s390x: big-endian byte order causes mass test verification failures

[ludomal]

Summary

395 out of 1337 tests (70% pass rate) fail on s390x (big-endian). The tools execute correctly but verify steps fail because output is in native big-endian byte order while reference test vectors are stored as little-endian.

Context

The STL already provides sb (src/unsup/sb.c) for byte-swapping between platforms, and the documentation states that files should be byte-swapped before processing on big-endian systems. However, the CTest infrastructure does not perform this step automatically.

Scope

Nearly every module is affected:

• Codecs: g711, g711iplc, g722, g726, g727, rpeltp
• Filters: fir, filter, iir (pcmdemo, cirsdemo, c712demo)
• Noise/measurement: mnru, p50fbmnru, bs1770demo, sv56, esdru
• Utilities: truncate, astrip, fdelay, scaldemo, spdemo, freqresp, reverb
• New tools: wav_io, eid-amr, dlyerr_2_errpat
• basop: W_round64_L precision test (Portability issue on basop W_round32_s / W_round48_L / W_round64_L #189)

Resolution

Use sb to byte-swap test input before running and byte-swap output before verifying on big-endian hosts. This aligns with the existing documented workflow.

Context

Identified during STL2026 CI validation (PR #190). The s390x job was added specifically to detect big-endian portability issues. This is a pre-existing problem now made visible by CI. Non-blocking for STL2026 but should be resolved for the next update.