audio: source-ground the MT6739 call path and calibrate echo control

[forkwright]

Finding

Phase-07 planning asserted a direct modem-PCM-to-MT6357 call-audio path that bypasses the AP and froze an unexplained >-30 dB echo threshold, while the same plan still asks how the real call path and readiness work. The repository has no pinned MT6739 PCM/voice-route source or accepted AEC metric/setup.

Evidence

• Kanon Phase-07 PLAN scope/decision rows assert direct modem PCM ↔ codec hardware routing and an AEC threshold.
• Its open questions still ask how call-audio readiness/path works, and REQ-16 says M7 measurement should decide echo treatment.
• Thumos source models audio sessions/routes but contains no cited MT6739 modem voice-path register/firmware contract.
• audio: couple every microphone transition to the audit log, then qualify MT6357 I/O #399/fix(m7): replace invalid direct PMIC MMIO with a source-grounded PWRAP seam #862 own mic-audit coupling and PMIC transaction safety, not modem PCM topology or AEC calibration.

Why this matters

A wrong route assumption can send invalid PMIC/AFE commands or leave calls silent. An unqualified dB threshold has no defined metric, setup, error bound, or intelligibility outcome and cannot drive a safe product decision.

Done when

• Pin authoritative MT6739/MT6357/BSP sources for modem voice PCM/AFE routing, ownership, clocking, sample format, readiness, teardown, and error reporting.
• Represent the route behind a host-testable seam and fail closed until every required endpoint acknowledges.
• Define AEC/echo metrics, fixtures, corpus, measurement setup, latency/intelligibility bounds, and versioned acceptance criteria without a guessed constant.
• Test call setup/teardown, route conflict, partial failure, timeout, mute, preemption, and unavailable codec/modem paths.
• Only after source/software acceptance, record operator-owned M7 voice and acoustic receipts.