Users of conda/mamba environments with often use nb_conda_kernels to keep notebook server and extensions consolidated to a single environment, and access various language kernels from other environments on the machine. This is to prevent needing so many separate jupyter client instances running (one for each environment), instead letting a single client "see" all the environments that have one or more language kernels installed.

Compatibility with nb_conda_kernels was already solved in #540, but looking at what exactly the enabling dependency spec coconut[jupyter] needs, from this section of coconut.constants, this will install another instance of jupyterlab to the current conda environment, which is adding a lot of overhead when the kernel should be simply showing itself to the separate instance of jupyter notebook/lab in its own environment.

• Can we remove the dependency on jupyterlab and others when installing the "coconut kernel" do reduce overhead for these environment installs? I would like to be able to use coconut in jupyter without reinstalling jupyterlab each time (i.e. via nb_conda_kernels)
• I believe a more common practice is to have a separate repository/package to only provide the kernel to the language, e.g. coco_kernel. Here's an example of the dependency needs for octave_kernel, which is using the metakernel tool to do a lot of the heavy lifting. Would you be interested in a separate coco-kernel repo based on metakernel to lighten up the dependency of coconut kernels for jupyter?