Break USB capture churn with backoff; instrument iso retire reason

[patrickrb]

Problem (from the crashing device)

Pulled debug.log + logcat from a Pixel 8 running with a C-Media USB adapter (0D8C:0012). Two field failures, both rooted in the libusb-direct capture path:

1. Waterfall freeze — USB capture thrashes. Each libusb capture session delivered ~100 ms of audio, then retired all its isochronous IN transfers (code=0), and MicRecorder restarted it. Over one session: 127 opens, 88 stop/restart cycles, every ~2 s. MicRecorder only counted a session as failed when it saw no data at all — but here a few samples always arrived first (sawData=true), so the safety net never tripped and it churned forever. Result: capture is alive only ~100 ms out of every 2000 ms (~5% duty cycle) → the waterfall barely updates and the decoder never gets a continuous window.

2. Native SIGSEGV in libusb_handle_events. That 2 s libusb_init/libusb_exit churn (plus a leaked context per cycle, and concurrent TX contexts) races libusb global state:

signal 11 (SIGSEGV) ... in libft8af.so
  #06 libusb_handle_events_timeout_completed+256

Fix

• UsbCaptureRetryPolicy (new, pure-logic, unit-tested): a session counts as a failure if it saw no data or stayed alive too briefly to be useful (< 5 s, under a third of a 15 s FT8 cycle). Reinit now uses exponential, capped backoff (2 s → 60 s) instead of a flat 2 s, and never silently falls back to the phone's built-in mic (per operator preference — radio audio or none). This cuts the init/exit churn ~15–30× and greatly reduces the crash rate.
• MicRecorder tracks per-session alive time; the failure tally resets only when a session was actually useful (not on the first sample), and it won't re-arm after a real stop.

Instrumentation (to land the targeted native fix next)

The native ft8af_usb_capture logcat tag does not reliably surface in the field, so we couldn't read why the iso transfers retire. Now:

• usb_audio_capture.cpp records the first cause and passes it via onCaptureStopped(code): 1000+transfer_status, 2000+(-submit_err), 3000+(-handle_events_err), or 1 (retired, no terminal cause).
• UsbAudioDevice.describeCaptureStopCode() decodes it into debug.log, e.g. code=1005 (transfer terminal status NO_DEVICE).

The next on-device run will state exactly why capture retires — the data needed for the definitive native stability fix (and, separately, to stop leaking a libusb context per cycle).

Scope / honesty

This fixes the freeze and substantially reduces the native crash by cutting churn frequency, but the SIGSEGV's root fix depends on the retire reason this PR instruments for. Follow-up will use that to make capture stable (or degrade cleanly) and eliminate the per-cycle context leak.

Tests

UsbCaptureRetryPolicyTest (isFailure / exponential+capped backoff / overflow guard), UsbAudioWriteErrorTest (capture-stop-code decode + transferStatusName). :app:assembleDebug (native compiles, all ABIs) + :app:testDebugUnitTest green.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 21.57%. Comparing base (f23faaa) to head (12c8804).

Additional details and impacted files

@@             Coverage Diff              @@
##                dev     #433      +/-   ##
============================================
+ Coverage     21.47%   21.57%   +0.09%     
- Complexity      133      140       +7     
============================================
  Files           150      154       +4     
  Lines         19755    19945     +190     
  Branches       2909     2947      +38     
============================================
+ Hits           4243     4303      +60     
- Misses        15344    15474     +130     
  Partials        168      168              

Flag	Coverage Δ
android	12.42% <ø> (+0.26%)	⬆️
desktop	39.86% <ø> (ø)
ios	96.38% <ø> (ø)
native	9.93% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.
see 5 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.