fix(silero): bump onnxruntime-node to 1.24.3 to fix libc++abi mutex abort (#1375)

[sgzrov]

Summary

Closes #1375.

@livekit/agents-plugin-silero pins onnxruntime-node to 1.21.0 exactly. That version (and 1.21.x..1.23.x) has a static-destructor race that aborts the Node process on shutdown when LiveKit's tokio runtime threads are still alive at exit — a 100%-reproducible crash on macOS arm64 with a libc++abi mutex error.

The bug was fixed upstream in onnxruntime-node@1.24.1. This PR bumps the pin (and the matching onnxruntime-common devDep) to 1.24.3.

Bisect

Same minimal repro, only changing the onnxruntime-node version:

onnxruntime-node	Result
1.20.1	clean exit
1.21.0	crash (silero's pinned version)
1.21.1 / 1.22.0 / 1.23.0 / 1.23.2	crash
1.24.1	clean exit — fix landed here
1.24.2 / 1.24.3 / 1.25.1	clean exit

Native stack trace at the abort

frame #8:  libonnxruntime.1.21.0.dylib`__clang_call_terminate + 12
frame #9:  libonnxruntime.1.21.0.dylib`std::__1::unique_ptr<OrtEnv,
              std::__1::default_delete<OrtEnv>>::~unique_ptr() + 84
frame #10: libsystem_c.dylib`__cxa_finalize_ranges + 416
frame #11: libsystem_c.dylib`exit + 44
frame #12: node`node::Exit(node::ExitCode) + 12
frame #15: node`Builtins_CallApiCallbackGeneric + 184  ← process.exit(0) from JS

10 tokio-rt-worker threads from rtc-node.darwin-arm64.node are still alive at the moment of abort (parked in __psynch_cvwait/kevent). Full lldb trace and bisection in #1375.

Test plan

• pnpm install resolves cleanly (lockfile regenerated, only the silero workspace is affected)
• pnpm --filter @livekit/agents build succeeds
• pnpm --filter @livekit/agents-plugin-silero build succeeds
• Minimal repro from silero.VAD.load() causes libc++abi mutex abort on process exit (macOS arm64) #1375 (silero.VAD.load() + process.exit(0)) exits cleanly with status 0 against the new version
• pnpm -w format:write and pnpm -w lint:fix produce no new warnings (existing 120 warnings in agents/src/voice/generation_tools.test.ts are unrelated)

onnxruntime-node 1.21 → 1.24 is a minor bump within the same major. The exposed API surface used by silero/src/onnx_model.ts (InferenceSession.create, Tensor, run-options) is unchanged across this range, and the silero plugin's own build + types pass.

[changeset-bot]

🦋 Changeset detected

Latest commit: c609294

The changes in this PR will be included in the next version bump.

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 1 additional finding.

[sgzrov]

Upstream root cause confirmed.

The bisect range matches the documented upstream regression and fix:

• Upstream issue: microsoft/onnxruntime#24579 — "mutex issue at process exit on MacOS since v1.21."* — identical error string, identical regression at 1.21, identical destructor-order diagnosis.
• Upstream fix: microsoft/onnxruntime#26445 — "[node] Fix logging mutex crash at exit on macOS", merged 2025-10-31, included in onnxruntime-node 1.24.1. Description: "Now we don't let the destructor of OrtEnv to be called if the program exits unexpectedly." Files touched are exactly the OrtEnv lifecycle: js/node/src/ort_singleton_data.{cc,h} and inference_session_wrap.cc.

That matches the lldb trace in #1375 (frame ~unique_ptr<OrtEnv> in libonnxruntime.1.21.0.dylib) precisely. Bumping silero's onnxruntime-node pin from 1.21.0 → 1.24.3 picks up that upstream fix.