Support single-file plugins in online and offline catalogs #419

Summary

Allow DataLab Desktop's online and offline plugin catalogs to distribute historical single-file datalab_*.py plugins alongside wheel distributions, without requiring authors to turn them into packaging projects.

Documentation first

Clarify three independent concepts before extending the code:

• Source organization: a single module or a Python package/project.
• Distribution: directly loaded source, an installed .py module, or a wheel.
• Functional capabilities: processing, input/output, visualization and applications.

Both source forms use the same plugin API. A wheel can contain a single module; the generated layered project is not an API requirement. Update both English documentation and French translations, including the self-contained offline PDF guide.

Expected behavior

• Support mixed local folders, explicit catalog.json entries and mirrored online catalogs containing modules and wheels.
• Preserve existing wheels, their dependency management, source priorities and withdrawal policies.
• Accept unchanged historical files. Inspect syntax and statically available metadata without importing or executing the script; explicit catalog metadata supplies information that cannot be determined statically.
• Track unversioned local scripts by SHA-256, without inventing versions. An unchanged artifact is already installed; a different revision requires explicit replacement.
• Recognize modules previously installed from a file, persist known catalog/version information compatibly, and support install, update, reload and uninstall with consent and collision checks.
• Modules use only already available libraries: no additional dependency installation or implicit package-index lookup for this format.
• Extend public catalog validation, GitHub release-asset selection, mirrored downloads and the website. Published module releases have an explicit catalog version and accepted license.

Acceptance criteria

• Catalog inspection and publication never execute plugin code.
• Complete local catalogs install with no network access in Python and standalone Desktop editions.
• Generated, historical, unversioned and dynamically described scripts have predictable behavior.
• Invalid syntax, unsafe paths, excessive sizes, digest mismatches, collisions and failed writes are surfaced without activating partial replacements.
• English/French documentation clearly explains both forms and their installation/update workflows.

Delivery

Implement on a dedicated branch based on DataLab develop, independently of experimental data integration. Open a Desktop PR targeting develop and a linked companion PR in DataLab-Platform/plugins targeting main. DataLab-Web, per-plugin environments, releases and publication of new example plugins are out of scope.