Task updating: Handle new parameters with default values

[jluethi]

@rhornb & @adrtsc discovered an issue with the way we handle new parameters getting added to a task & the task update logic.

Currently, if a task update add a new variable darkfield : bool = True that wasn't set in the old version, the update mode in Fractal web automatically updates the task without error messages. When one download the workflow, that variable is unset in the workflow (somewhat expected based on old task parameter logic).
Fractal web then displays an unset boolean as False. This is fairly misleading, because that parameter defaults to True, thus when the task is run, darkfield=True will be used, even though darkfield=False is shown in the web interface.

This example can be reproduced by upgrading form the APX Apply BaSiCPy Illumination Models task version 0.4.0 to 0.4.4.

Task pre upgrade (version 0.4.0):

Task after upgrade from 0.4.0 to 0.44 (without any warnings):

Task when added directly as version 0.4.4:

A similar issue can also be observed when other parameters with default values (e.g. the "correct_by" variable in screenshots above) get added.

Content of the workflow download

{
  "name": "Update_testing",
  "task_list": [
    {
      "meta_non_parallel": null,
      "meta_parallel": {
        "cpus_per_task": 1,
        "mem": 3750
      },
      "args_non_parallel": null,
      "args_parallel": {
        "illumination_profiles_folder": "test",
        "input_ROI_table": "FOV_ROI_table",
        "overwrite_input": true,
        "suffix": "_illum_corr"
      },
      "type_filters": {},
      "task": {
        "pkg_name": "apx-fractal-task-collection",
        "version": "0.4.0",
        "name": "Apply BaSiCPy Illumination Models"
      }
    },
    {
      "meta_non_parallel": null,
      "meta_parallel": {
        "cpus_per_task": 1,
        "mem": 3750
      },
      "args_non_parallel": null,
      "args_parallel": {
        "illumination_profiles_folder": "Test",
        "input_ROI_table": "FOV_ROI_table",
        "overwrite_input": true,
        "suffix": "_illum_corr"
      },
      "type_filters": {},
      "task": {
        "pkg_name": "apx-fractal-task-collection",
        "version": "0.4.4",
        "name": "Apply BaSiCPy Illumination Models"
      }
    },
    {
      "meta_non_parallel": null,
      "meta_parallel": {
        "cpus_per_task": 1,
        "mem": 3750
      },
      "args_non_parallel": null,
      "args_parallel": {
        "illumination_profiles_folder": "test2",
        "correct_by": "channel label",
        "darkfield": true,
        "baseline": true,
        "input_ROI_table": "FOV_ROI_table",
        "overwrite_input": true,
        "suffix": "_illum_corr"
      },
      "type_filters": {},
      "task": {
        "pkg_name": "apx-fractal-task-collection",
        "version": "0.4.4",
        "name": "Apply BaSiCPy Illumination Models"
      }
    }
  ]
}

(when reproducing, one needs to save the task with required parameters before the workflow download has any content in the args)

[jluethi]

I think we should handle this better during upgrading. Our current logic still relies heavily on the idea that only some parameters would get set, but we've switched to always setting all parameters when more of the upgrade handling moved to the backend, right?
If, during upgrading, we'd add the newly added parameters with their defaults, that would work correctly, both for booleans and strings.

[jluethi]

Another way to reword this: Parameters with a default value shouldn't end up in a None state (=> greyed out) in regular Fractal web usage. And current update procedures sometimes get us to that state

[tcompa]

I'm making a small change as part of #2355, but it's not meant to fix this issue.

I think it has to do with the webclient logic in https://github.com/fractal-analytics-platform/fractal-web/blob/2939112ef3890463f63dad6d7e56da2153b6ad5c/src/lib/components/v2/workflow/VersionUpdate.svelte#L155-L179.
TBD

[tcompa]

Now tracked in fractal-analytics-platform/fractal-web#728