fix(testing): cannot-fail scan judges a line, not a statement, so a second assertion on the same line is missed #6218

This was generated by AI during work-loop.

Context

Found by the Codex review on PR #6216 (#6040 PR B). The testing plugin's cant-fail scanner judges oracles per line, not per statement. A C# line holding two statements, such as Assert.True(true); Assert.NotNull(x);, is classified as a whole: the inert first statement decides the line, so the weak second assertion is never reported as rule-weak-oracle. Main has the same miss, so PR B did not introduce it; fixing it needs per-statement splitting in the scanner, larger than that PR.

Proposed work

Split C# lines on top-level ; (outside strings, char literals and parentheses) before oracle classification, so each statement is judged on its own. Check whether the Python, JS and Go lanes share the same per-line assumption.

Acceptance criteria

• A bad fixture with Assert.True(true); Assert.NotNull(x); on one line reports both inert and weak-oracle; red on the current engine.
• A stay-quiet fixture with two strong assertions on one line reports nothing.
• Findings on existing tracked C# tests are unchanged except where intended.

References

• PR fix(testing): report xUnit always-true and typeof/nameof constant asserts as inert #6216 (Codex thread), feat(testing): flag C# typeof-only, construct-only and BDD cannot-fail tests in the audit #6040

Metadata

Filed by the work-loop lane (instance melo-worker-1) per the self-observation contract; not self-triaged.