fix: add mutex to VoiceActivityDetection to prevent race between generate() and unload() (#1056)

## Description

Add thread-safety to `VoiceActivityDetection` using the same mutex
pattern already established by `VisionModel`.

Without this, `BaseModel::unload()` can destroy `module_` on the JS
thread while `generate()` is still calling `forward()` on a worker
thread, causing `SIGILL` / `SIGSEGV` crashes.

### Introduces a breaking change?

- [x] No

### Type of change

- [x] Bug fix (change which fixes an issue)

### Tested on

- [x] Android

### Testing instructions

Use `useVAD({ model: FSMN_VAD })` with continuous audio input and
trigger rapid mount/unmount cycles. Confirm no crashes or deadlocks.

### Related issues

Fixes #1055

---------

Co-authored-by: Bartosz Hanc <bartosz.hanc02@gmail.com>

Author: radko93

packages/react-native-executorch/common/rnexecutorch/models/voice_activity_detection/VoiceActivityDetection.cpp

@@ -54,8 +54,14 @@ VoiceActivityDetection::preprocess(std::span<float> waveform) const {
   return frameBuffer;
 }
 
+void VoiceActivityDetection::unload() noexcept {
+  std::scoped_lock lock(inference_mutex_);
+  BaseModel::unload();
+}
+
 std::vector<types::Segment>
 VoiceActivityDetection::generate(std::span<float> waveform) const {
+  std::scoped_lock lock(inference_mutex_);
 
   auto windowedInput = preprocess(waveform);
   auto [chunksNumber, remainder] = std::div(

packages/react-native-executorch/common/rnexecutorch/models/voice_activity_detection/VoiceActivityDetection.h

@@ -5,6 +5,7 @@
 #include <executorch/extension/tensor/tensor.h>
 #include <executorch/extension/tensor/tensor_ptr.h>
 #include <executorch/runtime/core/evalue.h>
+#include <mutex>
 #include <span>
 
 #include "rnexecutorch/metaprogramming/ConstructorHelpers.h"
@@ -23,7 +24,11 @@ class VoiceActivityDetection : public BaseModel {
   [[nodiscard("Registered non-void function")]] std::vector<types::Segment>
   generate(std::span<float> waveform) const;
 
+  void unload() noexcept;
+
 private:
+  mutable std::mutex inference_mutex_;
+
   std::vector<std::array<float, constants::kPaddedWindowSize>>
   preprocess(std::span<float> waveform) const;
   std::vector<types::Segment> postprocess(const std::vector<float> &scores,