Add Jupyter notebook support with OCaml kernel

[cuihtlauac]

Summary

The devcontainer currently has no Jupyter support. Adding the ocaml-jupyter kernel would align well with the project's goal of zero-friction onboarding for tutorials and training sessions.

Prior art

KC Sivaramakrishnan used this stack to teach CS3100 (Paradigms of Programming) at IIT Madras:

• Blog post
• Course repo (F19), M20
• Docker image
• patricoferris/ocaml-teaching — extends KC's images with editors and nbgrader examples

Current state

Most prerequisites are already in the base image:

• `python3` and `python3-pip`
• System libraries needed by the OCaml kernel: `libgmp-dev`, `libffi-dev`, `zlib1g-dev`
• `merlin` (used by the kernel for tab completion)

What's needed

1. System dependency (`base/Dockerfile`)

• Add `libzmq5-dev` (ZeroMQ, for kernel-frontend communication)

2. OCaml kernel (`dev/Dockerfile`)

• `opam install jupyter` — pulls in zmq, lwt, yojson, cryptokit, etc.
• Run `ocaml-jupyter-opam-genspec` to generate the kernel spec
• Run `jupyter kernelspec install` to register it
• Register a kernel per switch so users can select between 5.4.0 and 5.4.0+tsan from the Jupyter notebook interface. Each switch needs its own kernelspec with a distinct display name (e.g., "OCaml 5.4.0", "OCaml 5.4.0+tsan").

3. Python-side Jupyter

• `pip install jupyterlab`

4. nbgrader (optional, for autograding)

• `pip install nbgrader` — Jupyter extension for creating, distributing, and auto-grading notebook assignments
• Supports hidden test cells and read-only solution cells
• Language-agnostic: works with OCaml cells out of the box
• Known issue: KC reported nbgrader bugs where failing cells were incorrectly awarded grades. TAs should manually verify autograded results.
• Reference: patricoferris/ocaml-teaching contains a working nbgrader example

5. RISE or equivalent (optional, for live presentations)

• `pip install notebook rise` (for classic notebook) or `pip install jupyterlab-rise` (for JupyterLab)
• Turns notebooks into live reveal.js slideshows — used by KC for lectures
• Known limitation: KC reported poor font sizing and no diagramming support

6. DevContainer config

• Forward port 8888 in `devcontainer.json`

7. Testing

• Add `test/test-jupyter.sh` to verify kernel registration, server startup, and cell execution

Open questions

• Install in both switches (5.4.0 and 5.4.0+tsan) or just the default? Lwt/zmq threading under TSan may produce false positives.
• Estimated image size increase: ~100-170 MB (~3-5% of current size)
• Estimated build time increase: ~5-10 min (zmq + lwt compilation)
• Should nbgrader and RISE be in the main image or left for tutorial-specific layers?