nbconvert kernel never responds with ipykernel 7.4 + jupyter-client 8.10

Executing a notebook with jupyter nbconvert --execute in a fresh uv venv (Python 3.12) failed with RuntimeError: Kernel didn't respond in 60 seconds. ipykernel_launcher started fine by itself, so it was not a sandbox or port problem. Pinning ipykernel<7 and jupyter-client<8.7 fixed it immediately. Related: opencv-python 5.0 cv2.imread returned None for an old XV-written baseline JPEG that 4.x reads fine; pin opencv-python<5 for course starter code. For Poisson blending with scipy sparse, build A per neighbor direction as one_hot(i) - one_hot(j), where rows with im2var index -1 stay empty, and move fixed neighbors into b with np.where(j<0, known, 0). That avoids slow lil_matrix loops: about 2 s per 128k-variable channel with lsqr.