Parse docstrings for nested argument models

[tcompa]

After we introduce the first nested task-argument model (see #386), we should also parse the nested-model arguments description. This will probably be a bit custom, but it can be done as part of the dev tools.

[tcompa]

0. Extract _get_args_descriptions and _include_args_descriptions_in_schema into a new module fractal_tasks_core.dev.lib_descriptions.py.

1. Hard-code a list of inner Pydantic models with their module, e.g.

OmeroChannel, defined in fractal_tasks_core.lib_channels
Window, defined in fractal_tasks_core.lib_channels
Channel, defined in fractal_tasks_core.tasks._input_models
NapariWorkflowsInput, defined in fractal_tasks_core.tasks._input_models
NapariWorkflowsOutput, defined in fractal_tasks_core.tasks._input_models
...

2. For each one of this model, run a function (similar to _get_args_descriptions) that extracts the docstring (via ast) and parses it (via docstring_parser). The goal is to have something similar to

descriptions = {
    "attribute1": "TBD",
    "attribute2": "This is my docstring...",
} 

for each object in the list.

3. Write and run a function similar to _include_args_descriptions_in_schema, that goes into the definitions block and injects the new descriptions of custom-model attributes.