Add kernelretsnoop and threadhist tools for CUDA kernel profiling

- Introduced `kernelretsnoop`, an eBPF-based tool to trace CUDA kernel thread exit timestamps, providing insights into thread execution times and performance bottlenecks.
- Implemented `threadhist`, a tool to analyze per-thread execution counts in CUDA kernels, helping to identify load imbalances and optimize thread configurations.
- Added example CUDA application (`vec_add.cu`) for testing both tools, demonstrating their usage in real scenarios.
- Created README files for both tools, detailing their functionality, usage, and examples.
- Included necessary build files and configurations for compiling and running the tools.

Author: yunwei37

example/gpu/README.md

@@ -48,8 +48,8 @@ The GPU support is built on the `nv_attach_impl` system (`attach/nv_attach_impl/
 Complete working examples with full source code, build instructions, and READMEs are available on GitHub:
 
 - **[cuda-counter](https://github.com/eunomia-bpf/bpftime/tree/master/example/gpu/cuda-counter)**: Basic probe/retprobe with timing measurements
-- **[cuda-counter-gpu-array](https://github.com/eunomia-bpf/bpftime/tree/master/example/gpu/cuda-counter-gpu-array)**: Per-thread counters using GPU array maps
-- **[cuda-counter-gpu-ringbuf](https://github.com/eunomia-bpf/bpftime/tree/master/example/gpu/cuda-counter-gpu-ringbuf)**: Event streaming with ringbuf maps
+- **[kernelretsnoop](https://github.com/eunomia-bpf/bpftime/tree/master/example/gpu/kernelretsnoop)**: Captures per-thread exit timestamps to detect thread divergence, memory access patterns, and warp scheduling issues
+- **[threadhist](https://github.com/eunomia-bpf/bpftime/tree/master/example/gpu/threadhist)**: Per-thread execution histogram using GPU array maps to detect workload imbalance
 - **[rocm-counter](https://github.com/eunomia-bpf/bpftime/tree/master/example/gpu/rocm-counter)**: AMD GPU instrumentation (experimental)
 
 Each example includes CUDA/ROCm application source, eBPF probe programs, Makefile, and detailed usage instructions.

example/gpu/cuda-counter-gpu-array/README.md

@@ -1,4 +1,4 @@
-/cuda_probe
+/kernelretsnoop
 /.output
 /victim*
 /vec_add.cpp

example/gpu/cuda-counter-gpu-ringbuf/README.md

@@ -21,7 +21,7 @@ INCLUDES := -I$(OUTPUT) -I../../../third_party/libbpf/include/uapi -I$(dir $(VML
 CFLAGS := -g -Wall
 ALL_LDFLAGS := $(LDFLAGS) $(EXTRA_LDFLAGS)
 
-APPS = cuda_probe # minimal minimal_legacy uprobe kprobe fentry usdt sockfilter tc ksyscall
+APPS = kernelretsnoop # minimal minimal_legacy uprobe kprobe fentry usdt sockfilter tc ksyscall
 
 CARGO ?= $(shell which cargo)
 ifeq ($(strip $(CARGO)),)