Limit version of IPython to < 9.17.0 to avoid import error during notebook validation (#1118)

Some of the notebooks use Kera's `plot_model` function. That function
has a check for whether graphviz is installed. If it is _not_ installed,
`plot_model` then tests whether "IPython.core.magics.namespace" is
present in `sys.modules` and does one of two things: if it's present,
`plot_model` prints a message; if it is not present, it raises an
`ImportError`. Unfortunately, starting in IPython 9.17.0, IPython no
longer eagerly imports `IPython.core.magics.*` modules at startup.
Unless Graphviz is installed separately in the run-time environment,
calling `plot_model` will raise an `ImportError`.

Our notebook checks are done using our script
`ci_validate_tutorials.sh`, and that script pip-installs IPython, but it
only sets a lower bound on the version number. As a result, when the
script executes in CI, it installs a version of IPython >= 9.17.0, which
leads `plot_model` to fail the test for "IPython.core.magics.namespace",
which leads it to raise `ImportError` in CI.

The simplest solution is to add an upper bound on the version number in
`ci_validate_tutorials.sh`, which is what this PR does. This will solve
the CI failures and also help any users who may encounter the same
problem when testing notebooks locally.

Author: mhucka

scripts/ci_validate_tutorials.sh

@@ -23,7 +23,7 @@ export TF_USE_LEGACY_KERAS=1
 pip install \
   "nbclient==0.6.5" \
   "jupyter-client==7.4.9" \
-  "ipython>=8.10.0" \
+  "ipython>=8.10.0,<9.17.0" \
   "ipykernel>=6.29.0"
 
 # Gymnasium pip package needed for the quantum reinforcement learning tutorial