DOC: adapts documentation for first-time users

[stsievert]

What does this PR implement?
This is my first time running Dask-CHTC. I've adapted the documentation for items I'd like to see in the documentation. So far, these are:

• Adds note about installing Jupyter before running
• Only show dask-chtc jupyter run lab instead of that and the unuseful jupyter lab.
• Renames "Example Notebook" to "Dask Cluster Creation."

Reference issues/PRs
Other issues that came up in this first usage are #25, #26 and #27.

[stsievert]

I did have one question on the "Using the start, status, and stop subcommands" subsection: why can't I run dask-chtc jupyter run lab through tmux? That's how I normally persist Jupyter sessions on remote servers.

[JoshKarpel]

A few nitpicks, and one thought about the already-running-job message, but otherwise LGTM

[JoshKarpel]

I did have one question on the "Using the start, status, and stop subcommands" subsection: why can't I run dask-chtc jupyter run lab through tmux? That's how I normally persist Jupyter sessions on remote servers.

Yep, that should work just fine. I want there to be a workflow that doesn't involve tmux for those who don't like it, but for those who do, tmux is a great way to make a detached session.

Do note that it won't be quite as protected as the fully-detached dask-chtc jupyter start job would be. CHTC admins do reserve the right to kill userland programs; if you accidentally start behaving badly in some part of your tmux, they might kill the whole thing as a safety measure.

[stsievert]

CHTC admins do reserve the right to kill userland programs;

How can I be a good user so CHTC admins don't kill my jobs? I'm interested in running jupyter lab instead of dask-chtc jupyter run lab in a tmux session for maybe a couple days. What other downsides come from running jupyter lab instead of dask-chtc jupyter run lab?

[JoshKarpel]

How can I be a good user so CHTC admins don't kill my jobs?

The main consideration is: try not to use large amounts of CPU, memory, or disk I/O on the submit node for longer than a few minutes at a time. The HTCondor daemons need those resources to serve other people's jobs. For a Dask workflow, this means making sure that you do all the heavy-duty computation through Dask, pulling only the minimal information back to the scheduler.

What other downsides come from running jupyter lab instead of dask-chtc jupyter run lab?

Right now, very few (for you; there are some incidental benefits for us). But we're hoping to be able to enforce resource limits by doing this, which will make CHTC admins much happier with this workflow. As far as I know, there shouldn't be a difference in the behavior of Jupyter once you've got it running - let me know if there is, so that we can fix it :)