phase 5: edge transforms, soleprint-ui rename, infra fixes

- pipeline edge transforms: stages can declare accepted_transforms,
  edges carry a transform dict, runner injects per-stage and nodes
  apply (e.g. invert_mask before edge detection); editable from UI
  via PUT /config/edge-transform
- rename mpr-ui-framework -> soleprint-ui (now an external package
  synced via .spr from /home/mariano/wdir/spr); add @vue-flow/core
  and uplot to detection-app so linked package resolves them
- Tiltfile guards kubectl context, k8s commands pin --context kind-mpr
- kind-config: gateway on hostPort 30080 (Caddy fronts mpr.local.ar)
- modelgen: pyproject.toml, .spr marker, dict default_factory support

core/detect/graph/runner.py

@@ -213,6 +213,13 @@ class PipelineRunner:
         self.config = config
         self.do_checkpoint = checkpoint
         self.stage_sequence = _flatten_config(config, start_from)
+        # Build edge transform lookup: {target_stage: {source_stage: transform_dict}}
+        self._edge_transforms: dict[str, dict[str, dict]] = {}
+        for edge in config.edges:
+            if edge.transform:
+                if edge.target not in self._edge_transforms:
+                    self._edge_transforms[edge.target] = {}
+                self._edge_transforms[edge.target][edge.source] = edge.transform
 
     def invoke(self, state: DetectState) -> DetectState:
         """Run the pipeline on the given state. Returns final state."""
@@ -224,6 +231,14 @@ class PipelineRunner:
             if check and check():
                 raise PipelineCancelled(f"Cancelled before {stage_name}")
 
+            # Inject edge transforms into state so the stage can read them.
+            # Compatible with LangGraph — just a state dict key.
+            transforms = self._edge_transforms.get(stage_name, {})
+            if transforms:
+                state["_edge_transforms"] = transforms
+            elif "_edge_transforms" in state:
+                del state["_edge_transforms"]
+
             # 2. Run node function
             node_fn = _NODE_FN_MAP.get(stage_name)
             if node_fn is None: