feat: display ML forecasts in Gradio UI

Add ensemble forecast visualisation to portfolio analysis results:
- Display 30-day price predictions with direction indicators
- Show model count and confidence information
- Include Phase 2.5 duration in performance metrics
- Add phase_2_5_duration_ms to AgentState model
- Format forecasts with emoji indicators (📈/📉/➡️)
- Display below key metrics and above LLM synthesis

app.py

@@ -207,12 +207,15 @@ def format_performance_metrics(final_state: AgentState) -> str:
     # Phase Execution Times
     phase_1_ms = final_state.get("phase_1_duration_ms", 0) or 0
     phase_2_ms = final_state.get("phase_2_duration_ms", 0) or 0
+    phase_2_5_ms = final_state.get("phase_2_5_duration_ms", 0) or 0
     phase_3_ms = final_state.get("phase_3_duration_ms", 0) or 0
-    total_ms = phase_1_ms + phase_2_ms + phase_3_ms
+    total_ms = phase_1_ms + phase_2_ms + phase_2_5_ms + phase_3_ms
 
     metrics_md += "**Execution Timeline:**\n\n"
     metrics_md += f"- Phase 1 (Data Collection): {phase_1_ms:,}ms\n"
     metrics_md += f"- Phase 2 (Computation): {phase_2_ms:,}ms\n"
+    if phase_2_5_ms > 0:
+        metrics_md += f"- Phase 2.5 (ML Predictions): {phase_2_5_ms:,}ms\n"
     metrics_md += f"- Phase 3 (LLM Synthesis): {phase_3_ms:,}ms\n"
     metrics_md += f"- **Total**: {total_ms:,}ms ({total_ms / 1000:.2f}s)\n\n"
 
@@ -304,6 +307,28 @@ def format_analysis_result(final_state: AgentState, holdings: List[Dict[str, Any
 - **CVaR (95%)**: ${cvar_95:,.2f}
 """
 
+    # Format ML forecasts if available
+    forecasts_md = ""
+    ensemble_forecasts = final_state.get("ensemble_forecasts", {})
+    if ensemble_forecasts:
+        forecasts_md = "### ML Price Forecasts (30-day)\n\n"
+        for ticker, forecast_data in ensemble_forecasts.items():
+            if isinstance(forecast_data, dict):
+                models_used = forecast_data.get("models_used", [])
+                num_models = len(models_used) if models_used else forecast_data.get("metadata", {}).get("num_models", "N/A")
+
+                # Get first and last prediction (current vs 30-day)
+                predictions = forecast_data.get("predictions", [])
+                if predictions and len(predictions) > 0:
+                    first_pred = float(predictions[0])
+                    last_pred = float(predictions[-1])
+                    change_pct = ((last_pred - first_pred) / first_pred * 100) if first_pred > 0 else 0
+
+                    direction = "📈" if change_pct > 0 else "📉" if change_pct < 0 else "➡️"
+                    forecasts_md += f"- **{ticker}**: {direction} {change_pct:+.1f}% (using {num_models} models)\n"
+
+        forecasts_md += "\n*Note: Forecasts combine Chronos foundation model with statistical baselines.*\n\n"
+
     # Get LLM synthesis and recommendations
     ai_synthesis = final_state.get("ai_synthesis", "Analysis completed successfully.")
     recommendations = final_state.get("recommendations", [])
@@ -322,6 +347,7 @@ def format_analysis_result(final_state: AgentState, holdings: List[Dict[str, Any
 
 {metrics_md}
 
+{forecasts_md}
 {ai_synthesis}
 
 {recs_md}

backend/agents/workflow.py

@@ -393,6 +393,7 @@ class PortfolioAnalysisWorkflow:
 
             # Track phase duration
             phase_duration_ms = int((time.perf_counter() - phase_start) * 1000)
+            state["phase_2_5_duration_ms"] = phase_duration_ms
 
             logger.info(
                 f"PHASE 2.5 COMPLETE: Generated forecasts for {len(ensemble_forecasts)} assets ({phase_duration_ms}ms)"

backend/models/agent_state.py

@@ -67,6 +67,7 @@ class AgentState(TypedDict):
     # Performance Metrics
     phase_1_duration_ms: Optional[int]
     phase_2_duration_ms: Optional[int]
+    phase_2_5_duration_ms: Optional[int]
     phase_3_duration_ms: Optional[int]
     llm_input_tokens: Optional[int]
     llm_output_tokens: Optional[int]