add model loading status, effects preset dropdown, clean up UI

Backend:
- Generation service reports 'loading_model' status only when model
  is not yet in memory, then 'generating' once inference starts
- Migrate hf_offline_patch.py from print() to logging module
- Update ADDING_TTS_ENGINES.md for post-refactor file paths

Frontend:
- HistoryTable shows 'Loading model...' vs 'Generating...' based on step
- FloatingGenerateBox: replace instruct toggle + inline effects editor
  with an effects preset dropdown (third dropdown after language and engine)
- Instruct UI removed for now (form field preserved for future models)
- Remove focus ring from Select component globally

app/src/lib/api/types.ts

@@ -73,7 +73,7 @@ export interface GenerationResponse {
   instruct?: string;
   engine?: string;
   model_size?: string;
-  status: 'generating' | 'completed' | 'failed';
+  status: 'loading_model' | 'generating' | 'completed' | 'failed';
   error?: string;
   is_favorited?: boolean;
   created_at: string;