Remove legacy config fields and their fallback logic

Completely removes deprecated configuration fields instead of just comments:

**Removed Config Fields:**
1. `DataConfig.test_transpose` - Use `inference.data.test_transpose` instead
2. `TestTimeAugmentationConfig.act` - Use `channel_activations` instead

**Removed Fallback Logic:**
1. build.py: Remove cfg.data.test_transpose fallback (lines 262-263)
   - Now only uses cfg.inference.data.test_transpose
2. lit_model.py: Remove single activation fallback (lines 331-348)
   - Now requires channel_activations for TTA

**Impact:**
- 3 files modified
- 24 lines removed
- Breaking change: Old configs using deprecated fields will no longer work
- Users must update to modern config structure

Author: claude

connectomics/config/hydra_config.py

@@ -392,7 +392,6 @@ class DataConfig:
         default_factory=list
     )  # Axis permutation for training data (e.g., [2,1,0] for xyz->zyx)
     val_transpose: List[int] = field(default_factory=list)  # Axis permutation for validation data
-    test_transpose: List[int] = field(default_factory=list)  # Axis permutation for test data
 
     # Dataset statistics (for auto-planning)
     target_spacing: Optional[List[float]] = None  # Target voxel spacing [z, y, x] in mm
@@ -866,7 +865,6 @@ class TestTimeAugmentationConfig:
     flip_axes: Any = (
         None  # TTA flip strategy: "all" (8 flips), null (no aug), or list like [[0], [1], [2]]
     )
-    act: Optional[str] = None  # Single activation for all channels: 'softmax', 'sigmoid', 'tanh', None
     channel_activations: Optional[List[Any]] = (
         None  # Per-channel activations: [[start_ch, end_ch, 'activation'], ...] e.g., [[0, 2, 'softmax'], [2, 3, 'sigmoid'], [3, 4, 'tanh']]
     )

connectomics/data/augment/build.py

@@ -257,17 +257,15 @@ def _build_eval_transforms_impl(
         if mode == "val":
             transpose_axes = cfg.data.val_transpose if cfg.data.val_transpose else []
         else:  # mode == "test"
-            # Check both data.test_transpose and inference.data.test_transpose
+            # Use inference.data.test_transpose
             transpose_axes = []
-            if cfg.data.test_transpose:
-                transpose_axes = cfg.data.test_transpose
             if (
                 hasattr(cfg, "inference")
                 and hasattr(cfg.inference, "data")
                 and hasattr(cfg.inference.data, "test_transpose")
                 and cfg.inference.data.test_transpose
             ):
-                transpose_axes = cfg.inference.data.test_transpose  # inference takes precedence
+                transpose_axes = cfg.inference.data.test_transpose
 
         transforms.append(
             LoadVolumed(keys=keys, transpose_axes=transpose_axes if transpose_axes else None)

connectomics/lightning/lit_model.py

@@ -328,24 +328,6 @@ def _apply_tta_preprocessing(self, tensor: torch.Tensor) -> torch.Tensor:
 
             # Concatenate all channels back together
             tensor = torch.cat(activated_channels, dim=1)
-        else:
-            # Fall back to single activation for all channels (old approach)
-            tta_act = getattr(self.cfg.inference.test_time_augmentation, 'act', None)
-            if tta_act is None:
-                tta_act = getattr(self.cfg.inference, 'output_act', None)
-
-            # Apply activation function
-            if tta_act == 'softmax':
-                tensor = torch.softmax(tensor, dim=1)
-            elif tta_act == 'sigmoid':
-                tensor = torch.sigmoid(tensor)
-            elif tta_act == 'tanh':
-                tensor = torch.tanh(tensor)
-            elif tta_act is not None and tta_act.lower() != 'none':
-                warnings.warn(
-                    f"Unknown TTA activation function '{tta_act}'. Supported: 'softmax', 'sigmoid', 'tanh', None",
-                    UserWarning,
-                )
 
         # Get TTA-specific channel selection or fall back to output_channel
         tta_channel = getattr(self.cfg.inference.test_time_augmentation, 'select_channel', None)