Implement compatibility with nb-conda-kernels

[xhejtman]

nb_conda_kernels lists kernelspecs from conda environments under its
own names and drops the originals: chatbook becomes
conda-base-chatbook and python3 becomes conda-base-py. Chatbook
assumed Jupyter's plain names. Every cell failed with "Chatbook backend
kernel 'conda-base-py' is not installed", although that was the backend
Settings offered, and a cell whose backend did resolve stayed at [*]
because the frontend did not treat the notebook as a Chatbook.

The Chatbook kernel now lists and starts backend kernels with the
server's kernelspec manager. The server describes that manager (class
and JSON-safe config) in the Chatbook kernel's environment, beside the
execution-mode cap, and the kernel builds the same manager from it.
The server resolves backend names for rule matching with its own
manager.

Chatbook is recognized by kernelspec language rather than name: in the
backend list, the execution-mode cap, the force-off policy, the
frontend's session detection, the backend dropdown, and New Chatbook,
which now opens the installed Chatbook kernelspec. Convert to Chatbook
still writes chatbook as the kernelspec name; JupyterLab falls back
to the only kernelspec in that language when the name is not installed.