Render datafactory models' feature descriptions in the catalogs job — install views-datafactory there (deferred from #483 → #474 → here)

[Polichinel]

#483 isolated the catalogs job's get_queryset() call so a model whose data client is not installed no longer aborts the job, and said the real fix — rendering those querysets — "belongs with #474". #474 (PR #497) added the Data Source column and row, which needs no client (read by AST), and deliberately did not touch the workflow. So today the 34 datafactory models' READMEs still say Feature Description: No description provided.

Do

.github/workflows/update_catalogs.yml: install views-datafactory (the models pin >=1.9.0,<2.0.0) next to views_pipeline_core==3.3.0, in one pip install line so the resolver sees both (installing it into an existing env step by step pulled numpy 2 and broke r2darts2/views-evaluation — #486, register note). Then update_readme.py's get_queryset() loads a datafactory descriptor and the README gets its feature list.

Check before merging

• the job's runtime and the wheel's dependency footprint (views-datafactory 1.13.0 requires numpy<3, pyarrow<20, views-frames<2; a datafactory fetch is NOT needed — get_queryset() only imports the module).
• tests/test_update_readme_survives_missing_data_client.py must stay green: the guard uses a client name that cannot exist, so it is unaffected by installing the real one.

Not for epic #488's definition of done.