Support `.ipynb` with `colab run`.

[ggprior]

Currently colab run --gpu T4 notebook.ipynb attempts to execute the notebook .json as Python:

$ colab run tiny.ipynb
[colab] Creating session 'run-8f06e6'...
[colab] Session READY (run-8f06e6). Executing tiny.ipynb...
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
/tmp/ipykernel_1065/1694935949.py in <cell line: 0>()
      5 {
      6  "cells": [
----> 7   {"cell_type": "code", "execution_count": null, "id": "a", ... "source": ["print('hello from cell 1')\n", "x = 21"]},
      8   {"cell_type": "code", "execution_count": null, "id": "b", ... "source": ["print('cell 2 result:', x*2)"]}
      9  ],
NameError: name 'null' is not defined
[colab] Stopping session 'run-8f06e6'...
[colab] Session terminated.
=== EXIT CODE: 1 ===

That was a slight gotcha for me -- in particular as the same will work for colab exec.
Would you be interested in a PR that adds support for .ipynb on the colab run command?

[keys-i]

Hi!
I’m new to the project and looking for a good first contribution. I’d be happy to take a stab at adding .ipynb support for colab run if that would be welcome.

My understanding is that colab run currently sends the notebook file contents through as Python source, so the raw notebook JSON gets executed directly. I’d probably start by detecting .ipynb files and executing only the code cells, matching the behavior users might expect from colab exec.

Before I open a PR, is there a preferred approach you’d like me to follow here?