Fix LlamaDemo default temperature, lifecycle cleanup, and LLM memory support

- Change DEFAULT_TEMPERATURE from 0.0 to 0.6 to prevent greedy decoding
  from hitting EOS token immediately on first generated token
- Add onCleared() lifecycle handler to ChatViewModel for proper Module
  resource cleanup when ViewModel is destroyed
- Update fbjni to 0.7.0 for main_exp AAR compatibility
- Enable largeHeap in AndroidManifest for large LLM model support

Authored with Claude.

Author: psiddh

llm/android/LlamaDemo/app/build.gradle.kts

@@ -268,7 +268,7 @@ dependencies {
     implementation("io.coil-kt:coil-compose:2.4.0")
     implementation("androidx.camera:camera-core:1.3.0")
     implementation("androidx.constraintlayout:constraintlayout:2.2.0")
-    implementation("com.facebook.fbjni:fbjni:0.5.1")
+    implementation("com.facebook.fbjni:fbjni:0.7.0")
     implementation("com.google.code.gson:gson:2.8.6")
     implementation("com.halilibo.compose-richtext:richtext-commonmark:1.0.0-alpha02")
     implementation("com.halilibo.compose-richtext:richtext-ui-material3:1.0.0-alpha02")

llm/android/LlamaDemo/app/src/main/AndroidManifest.xml

@@ -12,6 +12,7 @@
     <application
         android:name=".ETLogging"
         android:allowBackup="false"
+        android:largeHeap="true"
         android:dataExtractionRules="@xml/data_extraction_rules"
         android:fullBackupContent="@xml/backup_rules"
         android:icon="@drawable/logo"

llm/android/LlamaDemo/app/src/main/java/com/example/executorchllamademo/ModuleSettings.kt

@@ -153,7 +153,7 @@ data class ModuleSettings(
     }
 
     companion object {
-        const val DEFAULT_TEMPERATURE = 0.0
+        const val DEFAULT_TEMPERATURE = 0.6
         val DEFAULT_MODEL = ModelType.LLAMA_3
         val DEFAULT_BACKEND = BackendType.XNNPACK
     }

llm/android/LlamaDemo/app/src/main/java/com/example/executorchllamademo/ui/viewmodel/ChatViewModel.kt

@@ -96,6 +96,13 @@ class ChatViewModel(application: Application) : AndroidViewModel(application), L
     private val executor: Executor = Executors.newSingleThreadExecutor()
     private val contentResolver = application.contentResolver
 
+    override fun onCleared() {
+        super.onCleared()
+        module?.close()
+        loadedModules.forEach { (_, m) -> m.close() }
+        loadedModules.clear()
+    }
+
     init {
         // Check for clear chat history flag BEFORE loading saved messages
         val moduleSettings = demoSharedPreferences.getModuleSettings()