app/test_rag_system.py is not run in CI and has pre-existing bugs that make the suite error at setup

Noticed while working on #110 (PR #135): `.github/workflows/build.yaml` only runs `ruff check`/`ruff format --check` - no test step ever executes `app/test_rag_system.py` (or the other `test_*.py` files in `app/`).

The suite also has bugs that predate my change and would make it error out immediately if it were run:

• `setUpClass` calls `cls.rag_system.rebuild_embeddings()` with no argument, but the method requires a `knowledge_base` argument.
• `test_compute_document_scores` calls `self.rag_system.doc_about_embeddings()` as a function, but it's an attribute (a tensor), not a method.
• `test_cache_check_rebuild_embeddings_on_error` patches `self.rag_system.rebuild_embeddings` to a fake before capturing the "real" method to call from inside the fake, so `real_rebuild_embeddings` actually ends up bound to the fake itself - calling it recurses into itself indefinitely.

None of this blocks anything today because the suite simply never runs, but it means the tests provide no actual safety net. Worth fixing the bugs and wiring `pytest`/`unittest` into `build.yaml` in a follow-up.

Start with .github/workflows/build.yaml and app/test_rag_system.py. Run the test file locally to reproduce the setup and test errors, then inspect rebuild_embeddings, doc_about_embeddings, and the cache test's patching order. Done means the app tests run successfully in CI and provide a working safety net.