Add Docker support and fix benchmark/data pipeline bugs

Containerize the application with ROCm GPU support for AMD Radeon R9700:
- Add Dockerfile with PyTorch/PyG ROCm 5.6 wheels
- Add docker-compose.yml with dashboard, live-trading, and train services
- Add .dockerignore and .env.example for configuration

Fix benchmark script to use batch_size instead of num_stocks for variable
dimensions, and replace fragile partial model surgery with a standalone MLP
for memory estimation.

Fix data pipeline to skip dates with no next trading date instead of
fabricating zero returns.

Add slippage to paper broker, optional mark-to-market prices to broker
interface, and health check endpoint for container orchestration.

src/data/pipeline.py

@@ -673,8 +673,8 @@ class StockDataPipeline:
         """Return the next date that has price data for at least one ticker."""
         candidate = date + timedelta(days=1)
         for _ in range(7):
-            for ticker in self.price_data.values():
-                if not ticker.empty and candidate in ticker.index:
+            for price_df in self.price_data.values():
+                if not price_df.empty and candidate in price_df.index:
                     return candidate
             candidate += timedelta(days=1)
         return None
@@ -742,13 +742,19 @@ class StockDataPipeline:
             # Build graph edges
             edge_index, edge_weight = self._build_edges(node_tickers, date, pit_cache)
 
-            # Build targets: next trading day's return for each stock
+            # Build targets: next trading day's return for each stock.
+            # Skip dates where no next trading date exists (e.g. tail end of dataset)
+            # to avoid polluting training labels with fabricated zero returns.
             next_date = self._next_trading_date(date)
+            if next_date is None:
+                continue
+
             y_vals: List[float] = []
-            for ticker in node_tickers:
+            valid_seqs_for_y: List[np.ndarray] = []
+            valid_tickers_for_y: List[str] = []
+            for i, ticker in enumerate(node_tickers):
                 if (
-                    next_date is not None
-                    and ticker in self.price_data
+                    ticker in self.price_data
                     and next_date in self.price_data[ticker].index
                     and date in self.price_data[ticker].index
                 ):
@@ -757,9 +763,23 @@ class StockDataPipeline:
                         / self.price_data[ticker].loc[date]["Close"]
                         - 1
                     )
-                else:
-                    ret = 0.0
-                y_vals.append(ret)
+                    y_vals.append(ret)
+                    valid_seqs_for_y.append(node_sequences[i])
+                    valid_tickers_for_y.append(ticker)
+
+            if not y_vals:
+                continue
+
+            if len(valid_tickers_for_y) < len(node_tickers):
+                # Rebuild x and edges using only stocks with valid targets
+                x = torch.tensor(np.array(valid_seqs_for_y), dtype=torch.float32)
+                mean = x.mean(dim=0, keepdim=True)
+                std = x.std(dim=0, keepdim=True).clamp(min=1e-8)
+                x = (x - mean) / std
+                edge_index, edge_weight = self._build_edges(
+                    valid_tickers_for_y, date, pit_cache
+                )
+                node_tickers = valid_tickers_for_y
 
             y = torch.tensor(y_vals, dtype=torch.float32).unsqueeze(1)