Improve NRI test assertions and error matching

[AutuSnow]

Follow-up from #193.

The PR review raised two testing improvements for the NRI paths:


1. Avoid matching errors by string where possible.
2. Prefer stricter public-behavior assertions without inspecting internal data structures directly.
   
   This issue tracks improving the NRI unit tests so they remain precise while testing through observable behavior.
   
   Scope:

• Review NRI tests that match exact error strings.
• Replace string matching with typed errors, sentinel errors, helper predicates, or more stable assertions where appropriate.
• Look for places where tests were loosened to avoid checking internal state.
• Reintroduce stronger checks through public behavior when practical.
• Keep the tests focused; avoid turning unit tests into broad integration tests unless that is intentional.
  
  This is not required for Fail closed during NRI synchronize #193. It is follow-up test maintainability work, likely suitable for the v0.3.0 cycle.

[AutuSnow]

/assign