[PierreRaybaut]

Fixes #415.

Problem

The documented processing plugin example plugins/examples/datalab_custom_func.py, used by the Custom function tutorial and the plugin guide, no longer loaded on develop: registering sipi.Wrap1to1Func(weighted_average_denoise) with register_1_to_1 raised ValueError: 'weighted_average_denoise' is not a valid computation function, so the plugin added no menu. ComputingFeature only accepted functions decorated with Sigima's @computation_function(), and ccb81ef had moved the example from compute_1_to_1, which does not check the function, to register_1_to_1, which does. No test loaded the shipped examples in DataLab.

The regression only exists on develop: no released version is affected.

Fix

• ComputingFeature validates its function with _check_feature_function: a decorated Sigima computation function, or, for a 1-to-1 feature, a Wrap1to1Func instance of sigima.proc.signal or sigima.proc.image. Such a wrapper is picklable (as the process pool requires), named after the wrapped function and returns the expected object; Sigima documents it as usable in DataLab, and DataLab already detects the plugin origin of such wrappers.
• Other callables remain refused at registration, with a message naming both accepted forms; a wrapper registered for another pattern is refused too, since it produces one object from one object.
• The example, the tutorial and the migration guide are unchanged.

Behaviour change

• 1-to-1 features accept Wrap1to1Func wrappers, which were refused.
• The error message for other functions now names the accepted forms. It still starts with '<name>' is not a valid computation function.

Tests

• New test_shipped_plugin_examples_load_in_datalab (plugins_app_test.py): loads every plugin of plugins/examples in DataLab from a plugin directory, checks their stable IDs, the menu of the custom function example and the result of its processing on an image, then again after Reload plugins. It also fails if an example is added without being listed. It fails on develop with the error above.
• processor_registry_unit_test.py: wrapped NumPy functions registered as signal and image 1-to-1 features (name, implicit feature ID, lookup by function); refusal for the four other patterns; refusal of plain functions with the accepted forms.
• Regression: datalab/tests/backbone/plugins, datalab/tests/features/plugins, processor registry, feature identity and contract consistency tests, 243 passed.
• Ruff format and check; Pylint reports no new message on the changed modules (three existing messages in datalab/gui/processor/base.py).

Documentation and translations

• doc/features/advanced/plugins.rst: the functions processing plugins may register, with a link to the processing plugin example (new plugin_example_processing label).
• French translation of this entry only; the French documentation builds with -W.
• No release notes: the regression was never released.

Out of scope

• The single-file plugin generated by datalab-plugin create (Create single-file plugins with datalab-plugin create #413, Create single-file plugins with datalab-plugin create #414) registers a decorated wrapper function; once this is merged, Create single-file plugins with datalab-plugin create #414 will be rebased and simplified to register Wrap1to1Func(sample_processing) directly.