Tenferro Bug-Fix PR

Follow ai/contribution-workflows/bugfix-pr.md as the canonical workflow. Always read ai/contribution-workflows/repository-remediation.md too. When the user asks for a batch of related bug reports, repository-rule audit findings, or same-root-cause remediation in one PR, follow that remediation workflow.

Read both workflow files before editing code. Apply the scope gate before implementation.

Proceed only when the change fixes behavior already intended by current docs, tests, issue discussion, or API contracts. If the fix needs a new public API, operation family, backend, dependency, feature flag, architectural layer, or AD semantics change, stop and use the tenferro-issue-intake workflow instead.

Keep the interaction incremental:

1. Confirm bug-fix scope.
2. Collect or derive the reproducer and expected result.
3. Inspect the affected crate boundary and current tests.
4. Search for same-root-cause and same-pattern bugs in nearby modules and equivalent eager/traced/CPU/GPU/FFI paths.
5. Implement the smallest correct fix, or a coherent batch when related instances share the same contract and verification path.
6. Add or update a regression test when practical.
7. Propose or update a general audit/repository rule when it would prevent the same class of bug; before adding a new rule, inventory existing rules and merge overlapping guidance when possible.
8. For false positives, record the evidence and add a nearby source comment, rustdoc note, or source-contract test when the invariant is not obvious.
9. Verify with targeted checks and document any skipped checks.
10. Draft the PR body from .github/pull_request_template.md; for batches, preserve coherent commits and do not use squash merge.