Remove automatic creation of default dataset when creating a project

[tcompa]

The project-creation endpoint also takes care of creating a first dataset. I think this feature gives us little benefit, since the dataset is anyway empty and it will have to be handled again later (e.g. to change its name/type or add resources). Also: this could be the input dataset for a workflow, but the output dataset still needs to be created.

In the same spirit of #566, I propose to remove this feature. That is, a new project would be created without any dataset. The goal is to remove not-so-crucial branches, in order to:

• Simplify maintenance and documentation
• Make the addition of new features more transparent, instead of relying on some "hidden" feature.

Higher-level side note: there were good reasons for adding this kind of features in the first place, but IMO this is not the case any more after a few months of experience.

Any feedback? @jluethi @gusqgm

[jluethi]

Good discussion to have, I actually wanted to move in a separate direction. My observation during web demos was that there's still quite a few steps involved, e.g. going into the input dataset, modifying its name (optional), modifying its parameters (type image), adding resources. Creating an output dataset, modifying its parameters (type zarr), adding a resource. If we can reduce the number of steps needed here, that would be a big help.

As such, I was thinking about proposing:

1. The default dataset should be named "InputDataset"
2. The type of the "InputDataset" should be set to image
3. An "OutputDataset" should be automatically created
4. The type of the output dataset should be set to zarr

Currently, all workflows we have require such a setup. And we could reduce the required steps by the user to just: Add the resources to input & output datasets (2 steps instead of 6).

Potentially, going forward, we could have different "project types", e.g. a blank project comes without any dataset, a Yokogawa project comes with those datasets, a Zarr project comes with input as OME-Zarr and different things in that direction. But for the current setup and for quite a while forward

[jluethi]

After discussion: Fine if server API works that way. But we may expose this differently in the web interface

[tcompa]

I think that the previous comments in this issue and the upcoming work in fractal-web (see fractal-analytics-platform/fractal-web#299) all suggest that this issue is now ready for implementation.

A side-comment

This is in line with ongoing work on #826, as they both contribute to the high-level goal of keeping the CRUD endpoints as lightweight as possible and moving custom logic into dedicated endpoints. This may also make it simpler to create "complex" flows from fractal-web.

[tcompa]

To do:

• Remove automatic dataset creation from project-creation endpoint
• Fully remove default_dataset_name (e.g. from ProjectCreate)
• This probably requires fixing a lot of test here
• Update the CLI client accordingly, and fix its tests (ref Align with fractal-server new feature (project creation does not create a dataset) fractal-client#565)
• Update all fractal-demos examples (ref Align with new server feature (project creation not creating a dataset by default) fractal-demos#67).

[jluethi]

Will also require updating all the fractal-demos. Makes sense though, let's just keep track of that as well :)