Add 2D/3D convolution support with tests and benchmarks

[kwsp]

• Add ND convolution engine (FFTConvEngineND) with cached FFTW plans for 2D and 3D FFT-based convolution (full and same modes)
• Fix member init order in FFTConvBufferND to match declaration order
• Add C++ unit tests for 2D (test_fftconv2d.cpp) and 3D (test_fftconv3d.cpp)
• Add C++ benchmarks for 2D and 3D convolution in bench_fftconv.cpp
• Add pybind11 bindings for convolve2d/convolve3d (double and float)
• Export convolve2d, convolve2d_, convolve3d, convolve3d_ in Python package
• Add Python tests validating against scipy.signal.fftconvolve
• Add Python benchmark script (bench_nd.py) comparing against numpy/scipy
• Update test/CMakeLists.txt to include 2D/3D test sources