Improve error reporting and use pre-defined token

[krassowski]

#238 demonstrates that for an unknown reasons the Jupyter Server does not get setup correctly for some users. It is very difficult to debug currently, so I tried to address this in multiple ways:

• removing one of the modes of failures which was in retrieving the token (which if not found could lead to promise rejection and error screen state)
• displaying the actual error that led to the error screen (and encouraging reporting of these errors)
• allowing users to change environment with the selector introduced in Custom python environment support #317
• rewriting the error message which was incorrectly suggesting that it was due to server not being in path, where in reality there are several code paths which could lead to the promise rejection and that screen; instead it is now more accurately titled "Jupyter Server Initialization Failed"
• I also removed the "Install Jupyter" link, because the JupyterLab Desktop now ships with a builtin conda environment so it is redundant and it was directing users to rather not very useful generic Jupyter website (so they could end up installing Notebook and getting frustrated that it did not help).
• I added code for parsing the actual version of Jupyter Server that got started, but end up not using it for now. We should consider if we want to add a check on the version at this level too - it could help against some inconsistent environments - but it can wait for another PR.
• The server standard error output is now logged into the console. We may want to change this in the future (or make it optional, only for debugging) but it seems important for this stage of development.

I also restored the XSRF checks as I do not see any reason for having those disabled when the token is properly set up.

Tested by manually adding a reject(new Error('test')):

And selecting a non-Python path fusing the first button:

The environment selector works well too:

[krassowski]

The CI errors are not related to this PR, but to lack of peer dependency pinning. Yarn is installing the latest versions of peer dependencies which causes compilation issues. Merging #337 will temporarily solve this issue, but we should find a way to better locking of peer dependencies.

[mbektas]

Thank you!