Require cell permute type 2 when native GPU is enabled

Native GPU fixed-step integration diverges with the default
interleave_permute_type=0 (non-interleaved nrn_solve on device). HOC callers
that set pc.gpu_enable(1) and pc.gpu_backend("native") without
pc.optimize_node_order(2) therefore produced wrong spike rasters on large
models such as Traub 82894.

Mirror the Python gpu.enable path (which already applies permute 2) and
CoreNEURON's GPU guard: when gpu.enable and gpu.backend="native" are both
active, call nrn_optimize_node_order(2) if permute is not already 2.
Trigger from set_enable(true) and set_backend() so ordering of the two HOC
calls does not matter.

Add unit test with a stub nrn_optimize_node_order verifying permute 0 and 1
are upgraded to 2.

Author: Grok

src/neuron/gpu/config.cpp

@@ -1,5 +1,7 @@
 #include "neuron/gpu/config.hpp"
 
+#include "node_order_optim/node_order_optim.h"
+
 #include <algorithm>
 #include <cctype>
 #include <cstdio>
@@ -59,9 +61,32 @@ Backend backend() noexcept {
     return config().backend;
 }
 
+void ensure_native_gpu_cell_permute() noexcept {
+#if defined(NRN_ENABLE_GPU)
+    if (!enabled() || !backend_native()) {
+        return;
+    }
+    if (neuron::interleave_permute_type == 2) {
+        return;
+    }
+    int const previous = neuron::interleave_permute_type;
+    neuron::nrn_optimize_node_order(2);
+    if (previous != 2) {
+        fprintf(stderr,
+                "neuron::gpu: native GPU requires cell permute type 2 "
+                "(interleave_permute_type was %d); using permute 2\n",
+                previous);
+    }
+#else
+#endif
+}
+
 void set_enable(bool value) noexcept {
 #if defined(NRN_ENABLE_GPU)
     config().enable = value;
+    if (value) {
+        ensure_native_gpu_cell_permute();
+    }
 #else
     (void) value;
 #endif
@@ -70,6 +95,7 @@ void set_enable(bool value) noexcept {
 void set_backend(std::string_view name) {
 #if defined(NRN_ENABLE_GPU)
     config().backend = parse_backend(name);
+    ensure_native_gpu_cell_permute();
 #else
     (void) name;
 #endif

src/neuron/gpu/config.hpp

@@ -45,6 +45,13 @@ bool use_cuda_launcher() noexcept;
 /** One-time stderr notice when native GPU runs with pc.nthread() > 1. */
 void warn_native_gpu_multithread_policy() noexcept;
 
+/**
+ * Native GPU fixed-step requires interleaved Hines solve (permute type 2).
+ * Called when gpu.enable and gpu.backend="native" are both active; sets
+ * interleave_permute_type to 2 if it is not already.
+ */
+void ensure_native_gpu_cell_permute() noexcept;
+
 namespace detail {
 void reset_config_for_testing();
 void set_enable_for_testing(bool value);

test/unit_tests/gpu/config.cpp

@@ -2,6 +2,15 @@
 
 #include "neuron/gpu/config.hpp"
 
+namespace neuron {
+int interleave_permute_type = 0;
+
+int nrn_optimize_node_order(int type) {
+    interleave_permute_type = type;
+    return type;
+}
+}  // namespace neuron
+
 TEST_CASE("gpu config defaults", "[gpu][config]") {
 #if !defined(NRN_ENABLE_GPU)
     SKIP("NRN_ENABLE_GPU required");
@@ -14,3 +23,30 @@ TEST_CASE("gpu config defaults", "[gpu][config]") {
     CHECK(neuron::gpu::device_count() == 0);
 #endif
 }
+
+TEST_CASE("native GPU requires cell permute type 2", "[gpu][config]") {
+#if !defined(NRN_ENABLE_GPU)
+    SKIP("NRN_ENABLE_GPU required");
+#else
+    neuron::gpu::detail::reset_config_for_testing();
+    neuron::interleave_permute_type = 0;
+
+    neuron::gpu::set_backend("native");
+    CHECK(neuron::interleave_permute_type == 0);
+
+    neuron::gpu::set_enable(true);
+    CHECK(neuron::interleave_permute_type == 2);
+
+    neuron::gpu::detail::reset_config_for_testing();
+    neuron::interleave_permute_type = 1;
+    neuron::gpu::set_enable(true);
+    neuron::gpu::set_backend("native");
+    CHECK(neuron::interleave_permute_type == 2);
+
+    neuron::gpu::detail::reset_config_for_testing();
+    neuron::interleave_permute_type = 2;
+    neuron::gpu::set_enable(true);
+    neuron::gpu::set_backend("native");
+    CHECK(neuron::interleave_permute_type == 2);
+#endif
+}