feat: persistent-pipeline reuse — push_blocking, node introspection, stateful wrapper

Adds three pieces needed to build one KPN network and reuse it across many
replays/configs instead of tearing down and rebuilding per run:

- Channel<T>::push_blocking (+ IVariantChannel/VariantChannel forwarding):
  lossless backpressure push that waits for space instead of dropping when
  the ring is full. PyNode's run_loop now uses it so a downstream consumer
  lagging behind never silently drops a frame.
- PyNetwork::node_ptr / node_stats: raw node handle by name (for a binding
  to dynamic_cast to a concrete wrapper and call functor-specific runtime
  setters) and a per-node timing snapshot for profiling.
- ObjectVariantNodeWrapper: variant-node adapter for functors that need
  runtime-constructed state (a Config, a loaded gallery), mirroring
  VariantNodeWrapper's channel plumbing but backed by ObjectNode<Obj>.

Built and used downstream in scene-actor-extraction's sae_kpn Python replay
bindings for repeated threshold-sweep evaluation of the same pipeline.

include/kpn/python/bindings.hpp

@@ -217,6 +217,20 @@ public:
         return it->second;
     }
 
+    // Raw node handle by name — lets a binding dynamic_cast to a concrete wrapper
+    // type and call its functor's runtime setters (persistent-pipeline reuse).
+    VNode* node_ptr(const std::string& name) { return &node_at(name); }
+
+    // Per-node timing snapshot for profiling where a replay spends its time.
+    std::map<std::string, double> node_stats(const std::string& name) {
+        auto& n = node_at(name);
+        NodeSnapshot s = n.node_snapshot(name, 0.0);
+        return {{"frames", double(s.frames_processed)},
+                {"exec_ms", s.ema_exec_ms}, {"max_ms", s.max_exec_ms},
+                {"blocked_ms", s.total_blocked_ms}, {"fps", s.throughput_fps},
+                {"cpu_ms", s.total_cpu_ms}, {"cpu_util_pct", s.cpu_util_pct}};
+    }
+
 private:
     VNode& node_at(const std::string& name) {
         auto it = nodes_.find(name);
@@ -422,13 +436,17 @@ private:
 
                 for (std::size_t i = 0; i < out_channels_.size(); ++i) {
                     if (out_channels_[i])
-                        out_channels_[i]->push(std::move(outputs[i]));
+                        // Lossless: wait for space rather than drop. A dropped frame
+                        // silently corrupts a replay's score; backpressure just slows
+                        // the producer. (Was push() + "drop on overflow".)
+                        out_channels_[i]->push_blocking(std::move(outputs[i]));
                 }
 
             } catch (const ChannelClosedError&) {
                 break;
             } catch (const ChannelOverflowError&) {
-                // drop and continue
+                // no longer reachable with push_blocking, kept for safety
+                break;
             }
         }
     }
@@ -530,7 +548,8 @@ void register_py_network(nb::module_& m, const char* class_name = "Network") {
         .def("read",   &Net::read,
              nb::arg("node"), nb::arg("out_idx") = std::size_t(0))
         .def("write",  &Net::write,
-             nb::arg("node"), nb::arg("in_idx"), nb::arg("value"));
+             nb::arg("node"), nb::arg("in_idx"), nb::arg("value"))
+        .def("node_stats", &Net::node_stats, nb::arg("node"));
 }
 
 } // namespace kpn::python