Shim to support the notebook extensions

[echarles]

This PR add a shim to ensure the classic notebook extension installed on top of nbclassic do not break while importing notebook packages being no more available in nbclassic. To achieve this backwards compatibility, the needed module/package are shimmed, based on the existence or not of specific versions of notebook (V6, >V7).

• Discuss if we need to keep distinct shimming logic for notebook v6 and notebook v7.
• Discuss if we need to consider a more generic module loading shim based on https://docs.python.org/3/library/importlib.html#importlib.abc.Loader

[echarles]

For classic extensions to be operational, we may also need jupyter-server/jupyter_server#868 (add back IPythonHandler) and depend on a new jupyter-server release with that patch.

[echarles]

We have discussed this PR and the options to ensure backwards compatibility of the notebook extensions for nbclassic at last community call with @Zsailer and @afshin . I have further discussed the options with @ericsnekbytes , @RRosio and @seibert.

My last commit takes into account the feedback to not add IPythonHandler into jupyter-server. There are advantages to progress with the option of this PR to:

1. Have a single place where the shim occurs. To further adopt the idea to not inject shim in non-related repos like jupyter-server, it sounds good and in-line to also not inject shims in the notebook repos.
2. We should avoid allowing extensions developers of notebook 7 to use the old notebook packages (notebook.base..) as we want at some point to remove that shim. In other words, if we allow them to use eg. notebook.base..., those extensions will break when the shim that would be in notebook 7 will be removed (e.g. in notebook 8)

I will demo this PR at today community call with various setups (nbclassic with a working notebook extension while notebook 7, or notebook 7 or no notebook is installed).

cc/ @kevingoldsmith

[echarles]

@Zsailer We have just discussed the options at the community call and would love you feedback on the option proposed in this PR. I have given the demo and will be happy to replay it in one-one if you like.

We have decided to re-discuss this during next week community call. Of course, we don't need to wait that call to further brainstorm and exchange on this 👍

[Zsailer]

1. Have a single place where the shim occurs. To further adopt the idea to not inject shim in non-related repos like jupyter-server, it sounds good and in-line to also not inject shims in the notebook repos.
2. We should avoid allowing extensions developers of notebook 7 to use the old notebook packages (notebook.base..) as we want at some point to remove that shim. In other words, if we allow them to use eg. notebook.base..., those extensions will break when the shim that would be in notebook 7 will be removed (e.g. in notebook 8)

👍 Totally agree with these two goals.

I think this approach is good. Let's roll with it. If we hear reports of major issues coming from patching sys.modules, we can always change course. In the meantime, this is a nice, clean solution. 🚀

Great work, @echarles!

[echarles]

Super! Thx a log @Zsailer for the review and though with this. I see relevant CI failure due to the usage of assert. Let me fix that before merging 👍

[echarles]

The failing selenium tests are also encountered in #114. This is prolly a regression and not related to this PR.

I have fixed the relevant tests, so this is good to go. Merging.

[echarles]

I have opened #114 to follow-up on the selenium tests.