fix(ai): fix provider switching in wizard and remove model filtering

- Use local state for currentProvider in WelcomeWizard to ensure UI updates instantly
- Remove capability and keyword filtering for Gemini and Groq models in ProofreadService
- Add fetchAvailableModels stubs to offline/offlinelite flavors to fix build failures

Author: LeanBitLab

app/src/offline/java/helium314/keyboard/latin/utils/ProofreadService.kt

@@ -195,6 +195,8 @@ class ProofreadService(private val context: Context) {
     fun getProvider(): AIProvider = AIProvider.GROQ
     fun setProvider(provider: AIProvider) { /* No-op */ }
 
+    suspend fun fetchAvailableModels(provider: AIProvider): List<String> = emptyList()
+
     // API-compatible methods
     fun getApiKey(): String? = null
     fun setApiKey(apiKey: String?) { /* No-op */ }

app/src/offlinelite/java/helium314/keyboard/latin/utils/ProofreadService.kt

@@ -20,6 +20,8 @@ class ProofreadService(private val context: Context) {
     fun getProvider(): AIProvider = AIProvider.GEMINI
     fun setProvider(provider: AIProvider) { /* No-op */ }
 
+    suspend fun fetchAvailableModels(provider: AIProvider): List<String> = emptyList()
+
     fun getApiKey(): String? = null
     fun setApiKey(apiKey: String?) { /* No-op */ }
     fun hasApiKey(): Boolean = false