test(coverage): 11 test functions whose bodies execute nothing, counted for the first time #15256

Measured for the first time by #15189's third sweep. The shape was never counted — not because detecting it was hard, but because nobody had asked.

What

11 test functions whose bodies execute nothing — every statement is pass, ..., a bare string, or a bare print(...) — while the function name asserts a behaviour:

All eleven are in autobot-backend; every other tree measures zero.

Why neither existing detector sees them

• The returning-test guard looks for a return — these do not return.
• The swallowed-assertion guard looks for an assert under a bare handler — these have no assert at all.

So a function with no body passes both guards and shows green. The known worst instance of this shape was test_migrated_files_import, which had a body of pass plus six print() calls announcing six successful imports — the import statements having been stripped at some point. It asserted six imports while performing none, and reported success for all six.

PR #15254 adds detection (74 lines, driven off the existing collection model rather than a second AST walk) and ratchets the population at 11, down-only. It deliberately excludes skip/skipif/xfail — a declared skip is honest — and anything that calls something, which is a smoke test rather than an empty one.

What remains

The eleven are now counted and pinned, not fixed. Each needs a real test written, and each needs something the repository does not currently have on hand: the rag_integration and knowledge_grounding ones need a live endpoint or a real KnowledgeBase.

That is the honest position — and it is why they were ratcheted rather than deleted. Deleting them would drop the record that someone intended coverage here; leaving them unmarked would keep eleven green tiles that assert nothing.

Acceptance criteria

• For each of the eleven, decide: write the test, or delete it and record what was intended to be covered. A third option — leave it empty and green — is what this issue exists to end.
• Any that stay pending get an explicit pytest.mark.skip with a reason and an issue reference, so the guard's honest-skip exclusion applies and the tile stops reading as a pass.
• The ratchet goes down as each is resolved, never up.
• For each written test: a contrast mutation showing it fails when the behaviour it names is broken.
• Confirm no twelfth has appeared since — the guard now makes that a CI failure rather than a discovery.

Refs #15189, PR #15254.