feat(jieli): AC792 audio board wiring, adapter contract tests, and the JIELI pin bump

[maidang-xing]

Companion to tuya/TuyaOpen-JieLi#11. Depends on that PR merging first — the pin below points at a commit on that PR's branch, so a build of this branch before #11 merges will fail to resolve the platform commit.

What this carries

• boards/JIELI/AC792N_Develop_Board/audio_config.h — records which onboard microphone is captured and why (ADC1 over ADC0, measured), and that the two microphones cannot be opened together because the vendor ADC asserts on the second one.
• boards/JIELI/AC792N_Develop_Board/CMakeLists.txt — exports JIELI_BOARD_AUDIO_CONFIG so the CMake adapter target compiles the audio sources against this board's mic-routing macros, matching what the staged vendor Makefile already does. Without it the wl83 leg does not build at all.
• tests/platform/test_jieli_tkl_audio_output.py — regression tests pinning the playback volume path (the vendor state switch, the 0-100 DAC level scale, the Q14 digital-gain curve and its ordering after channel start) and the capture path (the ADC 0-19 gain scale, which is a different range from the DAC's and must not be "aligned" to it).
• platform/platform_config.yaml — JIELI pin moves from ec4c6cce to 8e5fdf3d, the head of the platform branch.

The pin move exposed stale platform contracts — they are repointed here

Moving the pin past the TKL adapter refactor turned 35 of this repo's source-contract tests red. Those tests read platform source files as text and assert on their shape, and the refactor moved the code they were written against — so they had been stale all along, latent only because the pin had not moved.

They are repointed, not weakened; each assertion now pins the mechanism that replaced the one it described:

• the audio provider tests scope to the WL83 region of the folded tkl_audio.c;
• the build tests follow the logic from jieli_build.py into tools/jieli_build/;
• the entry-point assertions follow tuyaos_switch_app_main.c to tuyaos/entry/jieli_app_entry.c;
• the backend-install assertions invert to pin the vtable's absence, which is the point of the fold;
• the queue tests follow the vendor os_q_* API to the local ring buffer that replaced it (chosen because the two SDKs disagree on os_q_post_to_back()'s message semantics and os_q_pend()'s timeout encoding);
• the K1-key assertions follow the build-time board.c patching to the board adkey config and TDL registration that replaced it;
• the MAC assertions follow the deleted tkl_jieli_chip_mac.c into tkl_wifi.c / tkl_bluetooth.c.

One test was dropped rather than repointed: test_jieli_uart_hello.py asserted a bring-up banner that the refactor deleted from tuyaos_app_main.c, and read examples/get-started/jieli_uart_hello, which has never been committed to this branch (it lives on feat/jieli-7916-uart). It had failed since before the pin move. It should come back with the example that gives it meaning.

Verification

• Outer repo: 83 passed, 10 subtests passed.
• Platform repo: 40 passed, 4 subtests passed.
• Both chips build green against the platform branch (switch_demo on wl83 and wl82, output_speaker on wl83 — BUILD SUCCESS, zero undefined references).
• Hardware-verified on an AC792N board: the 1 kHz boot tone plays and the ADKEY microphone loopback returns the speaker's voice.

Note

The two audio fixes themselves (playback audibility, capture gain) are diagnosed and documented in tuya/TuyaOpen-JieLi#11, where the adapter code lives.

test_jieli_config.py now pins the platform registry commit SHA literally, so it needs a one-line update on every future pin move. That is deliberate — the drift guard is what caught this one.

---

Added since first opened

The pin moved again, to 7d24da28 — the platform branch now also carries the board/TKL/app layer separation (see tuya/TuyaOpen-JieLi#11), which had not been pushed when this PR first went up. The registry drift guard in test_jieli_config.py moved with it.

your_chat_bot gains the gitignored-license hook the other apps have. switch_demo and friends keep real credentials in a gitignored tuya_config_secrets.h that tuya_config.h pulls in via #if __has_include; this app had none, so its tracked header was the only place to put a license. Same hook added, placeholders guarded, plus a tracked .example template.

A crash fix, and an honest caveat about it. your_chat_bot crashed and rebooted ~0.47 s into startup on AC792N, at the instant wq_highpri started. A controlled A/B — same app, same board, firmware built from the pre-change platform commit — crashed identically, so the defect predates the layer separation. The evidence points at the work-queue stacks: switch_demo carries CONFIG_STACK_SIZE_MSG_QUEUE=10240 / CONFIG_STACK_SIZE_WORK_QUEUE=8192 with a comment recording that the 4096 default overflowed wq_highpri and corrupted FreeRTOS list nodes; your_chat_bot carried neither. Confirmed on hardware. Flashed, and the app now passes the old crash point and runs on: reset_netcfg's timer starts, netmgr [wifi] register start, tuya_iot_init success at 1.391 s (the old crash was at 0.472 s), and it is still alive at 11.95 s with no reboot in the capture. Note commit a05dfaea's body still carries the pre-verification caveat; it is superseded by this, and the history is left unrewritten rather than force-pushed.

One test repointed. test_full_stack_audio_provider_is_registered_before_tkl_init pinned a call that the layer separation moved into tkl_init(); it had been failing unnoticed because no step between those commits ran the suite. Now pins the new location at equal strength.

Outer repo suite: 83 passed, 10 subtests passed. Platform suite: 40 passed, 4 subtests passed.

[maidang-xing]

Superseded by #730, which carries the same work squashed onto xb/tuyaopen-jl.

This PR was based on master; the Jieli work belongs on xb/tuyaopen-jl (#725 and #715 already target it). #730 also adds the AC792N audio board wiring, the your_chat_bot stack fixes and the adapter contract tests, and moves the JIELI pin to a84e800c (the platform commit merged as tuya/TuyaOpen-JieLi#11) — the old pin 7d24da28 is no longer reachable on the platform's master.