VSCode Quarto extension interactive window can't start a process-based dask LocalCluster/Client

I need to run the dask client/local-cluster with processes rather than threads, because it seems to do a much better job of memory management that way, and long running jobs can finish without running out of memory, but if I create the following qmd file:

```{python}
from dask.distributed import Client
```

Creating a cluster with processes=False:

```{python}
client = Client(processes=False)
display(client)
```

Creating a cluster with processes=True:

```{python}
client = Client(processes=True)
display(client)
```

Then when I step through the cells with [Shift]+[Enter], the first example (using threads) runs ok in the interactive window, but the second example (using processes) prints out a massive set of repeating warnings and errors before eventually failing with RuntimeError: Nanny failed to start.

At the very top of these errors it says (many many many times):

Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/usr/local/lib/python3.12/multiprocessing/spawn.py", line 122, in spawn_main
    exitcode = _main(fd, parent_sentinel)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/multiprocessing/spawn.py", line 131, in _main
    prepare(preparation_data)
  File "/usr/local/lib/python3.12/multiprocessing/spawn.py", line 246, in prepare
    _fixup_main_from_path(data['init_main_from_path'])
  File "/usr/local/lib/python3.12/multiprocessing/spawn.py", line 297, in _fixup_main_from_path
    main_content = runpy.run_path(main_path,
                   ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<frozen runpy>", line 286, in run_path
  File "<frozen runpy>", line 259, in _get_code_from_file
  File "/workspaces/dask_with_qmd_example/minimal-cluster-example.qmd", line 1
    ```{python}
    ^
SyntaxError: invalid syntax
...

Suggesting that it is trying to read the qmd file in as a python file and failing (understandably).

It works file when rendering the document using quarto render minimal-cluster-example.qmd --to html

It also works fine with similar code with a bare python file at the command line, and when executing that file line by line in a VSCode REPL or Jupyter Interactive Window. You do need to put if __name__ == "__main": around the script when running at the command line though.

It also works fine with a similar set of code in a Jupyter Notebook running interactively inside of VSCode.

If you break up the code into creating a LocalCluster and then linking the Client, it is linking the client that causes the problem, not creating the cluster. It looks a lot like something is trying to import the local python module, and failing because it is getting a quarto file instead of a python file (which is similar to the reason you need the if __name__ == "__main__": block in the bare python I think.

Not being able to run this interactively is making it a lot harder to build up to the final analysis that I can run in one go with quarto render.

Thanks for your help and/or any assistance or workarounds that might be possible.

Reproduction Information

VSCode version 1.96.1 running on Windows 10 Enterprise. Connected to a dev container with the following configuration (using Podman 5.3.1):

{
	"name": "Python 3",

	"image": "mcr.microsoft.com/devcontainers/python:1-3.12-bullseye",
	"features": {
		"ghcr.io/rocker-org/devcontainer-features/quarto-cli:1": {}
	},

	"postCreateCommand": "pip3 install --user -r requirements.txt",

	// https://medium.com/@guillem.riera/making-visual-studio-code-devcontainer-work-properly-on-rootless-podman-8d9ddc368b30
	"runArgs": [
		"--userns=keep-id:uid=1000,gid=1000", 
	   ],
	"containerUser": "vscode",
	"updateRemoteUserUID": true,

	"customizations": {
		"vscode": {
			"extensions": [
				"ms-toolsai.jupyter"
			]
		}
	},
}

requirements.txt:

dask[distributed]
nbformat
nbclient
jupyter

Reproduce the failure in the VSCode interactive window using minimal-cluster-example.qmd and the requirements.txt dependencies, then compare Client(processes=True) with the working processes=False case. Trace how the qmd cell is launched and confirm that a process-based Dask Client no longer tries to parse the Quarto document as Python; verify that interactive execution and quarto render both still work.