jetson compile fix

Makefile

@@ -18,15 +18,23 @@ GPR_CPP_DIR ?= $(HOME)/Documents/VNA_AlGo/gpr_cpp
 GPR_CPP_LIB ?= $(GPR_CPP_DIR)/build/libgpr_core_cpp.a
 GPR_VENV ?= $(GPR_CPP_DIR)/../.venv
 GPR_CUDA_INC_DIR ?= $(dir $(firstword $(wildcard $(GPR_VENV)/lib/python*/site-packages/nvidia/*/include/nvrtc.h)))
+# cuda.h can live in a DIFFERENT pip package than nvrtc.h (Jetson: cuda_runtime
+# vs cuda_nvrtc); the unified cu13 package on dev machines has both in one dir.
+GPR_CUDA_RT_INC_DIR ?= $(dir $(firstword $(wildcard $(GPR_VENV)/lib/python*/site-packages/nvidia/*/include/cuda.h)))
 GPR_NVRTC_LIB ?= $(firstword $(wildcard $(GPR_VENV)/lib/python*/site-packages/nvidia/*/lib/libnvrtc.so*))
 # System-CUDA fallback. An empty GPR_CUDA_INC_DIR would also corrupt the compile
 # command: a bare `-I` swallows the next argument.
 ifeq ($(strip $(GPR_CUDA_INC_DIR)),)
 GPR_CUDA_INC_DIR := /usr/local/cuda/include
 endif
+ifeq ($(strip $(GPR_CUDA_RT_INC_DIR)),)
+GPR_CUDA_RT_INC_DIR := /usr/local/cuda/include
+endif
 ifeq ($(strip $(GPR_NVRTC_LIB)),)
 GPR_NVRTC_LIB := $(firstword $(wildcard /usr/local/cuda/lib64/libnvrtc.so*))
 endif
+# Deduplicated CUDA include flags (a single dir on dev machines).
+GPR_CUDA_INC_FLAGS := $(addprefix -I,$(sort $(GPR_CUDA_INC_DIR) $(GPR_CUDA_RT_INC_DIR)))
 GPR_CUDA_LIB_DIR := $(dir $(GPR_NVRTC_LIB))
 # libnvrtc from pip ships only as libnvrtc.so.N (no .so symlink), so link the
 # full path; the rpath makes the binary find it without LD_LIBRARY_PATH.
@@ -34,7 +42,7 @@ DATA_PROCESSOR_LDFLAGS := $(LDFLAGS) $(GPR_CPP_LIB) $(GPR_NVRTC_LIB) -lcuda -Wl,
 
 INCLUDES := \
 	-I$(GPR_CPP_DIR)/include \
-	-I$(GPR_CUDA_INC_DIR) \
+	$(GPR_CUDA_INC_FLAGS) \
 	-Idata_acq_and_processing/common_cpp/ipc/include \
 	-Idata_acq_and_processing/common_cpp/config/include \
 	-Idata_acq_and_processing/third_party \