Clarification on HILT Scoping in Dependency Injection: Networking and ViewModel Components

I've been exploring HILT documentation and came across the information that scoping a binding incurs costs in both generated code size and runtime performance. The general guideline is to scope a binding only if it's necessary for the correctness of the code.

I'm seeking clarification on when it is appropriate to scope dependencies. To illustrate, I have two examples:

Example 1: Networking Retrofit Dependencies

I have my networking Retrofit dependencies installed in SingletonComponent. Should I consider scoping them, and if so, what factors should I take into account?

@InstallIn(SingletonComponent::class)@Moduleobject RetrofitModule {    private const val BASE_URL = "xyz"    @Provides    fun provideLoggingInterceptor(): HttpLoggingInterceptor = HttpLoggingInterceptor().apply {        level = HttpLoggingInterceptor.Level.BODY    }    @Provides    fun provideOkHttpClient(loggingInterceptor: HttpLoggingInterceptor): OkHttpClient =        OkHttpClient.Builder()            .addInterceptor(interceptor = loggingInterceptor)            .build()    @Provides    fun provideRetrofitInstance(okHttpClient: OkHttpClient): Retrofit = Retrofit.Builder()        .baseUrl(BASE_URL)        .client(okHttpClient)        .addConverterFactory(GsonConverterFactory.create())        .build()}

Example 2: Mappers and Repository Dependencies in ViewModelComponent

For mappers and repository dependencies installed in ViewModelComponent, should scoping be applied? What considerations are important in this context?

@InstallIn(ViewModelComponent::class)@Moduleabstract class DataModule {    companion object {        @Provides        fun provideCountMapper(): CountMapper = CountMapper()        @Provides        fun provideCountListMapper(countMapper: CountMapper): CountListMapper =            CountListMapper(countMapper = countMapper)         @Provides        fun providesCoroutineDispatcher(): CoroutineDispatcher = Dispatchers.IO     }    @Binds    abstract fun bindXYZRepository(xyzRepositoryImpl: XYZRepositoryImpl): XYZRepository}

Additionally, is it accurate to state that scoping a stateless object generally doesn't provide significant benefits and may introduce inefficiencies? Conversely, scoping a stateful object ensures consistent behavior and data management within its lifecycle.