MCPJVM-386: Infinite-loop prevention and terminal status classification for long-running execution

[nimbly-dev]

Description

A resilient orchestrator must stop deterministically when a long-running workflow never converges. This story adds bounded stop enforcement and explicit terminal classification for resumed long-running execution.

Scope

In scope:

• enforce bounded stop behavior from required project defaults
• prevent infinite resume/poll loops
• define deterministic terminal reason codes and final statuses for long-running resumed execution
• distinguish active/in-progress, timed-out, fail-closed, and completed-after-resume outcomes clearly

Out of scope:

• operator-driven repeated rerun strategy
• watcher/provider contract redesign
• full result-rendering redesign outside orchestration status needs

Acceptance Criteria

• Orchestrator stops deterministically when configured outer poll limits are exhausted
• Orchestrator stops deterministically when configured outer timeout budget is exhausted
• Final states and reason codes are explicit for bounded-stop outcomes
• Infinite resume loops are prevented by required defaults
• Terminal classification remains deterministic across resumed long-running execution

Implementation Notes

• Expected reason-code direction includes explicit outer-loop reasons such as poll exhaustion, timeout exhaustion, and progress-stalled classification where applicable.
• Preserve the distinction between business-state failure and outer orchestration stop reason where both are relevant.

Testing

• UT for bounded-stop rule evaluation and terminal reason-code mapping
• UT for infinite-loop prevention logic
• IT for a non-converging watcher/external-verification workflow that is stopped by required orchestrator defaults rather than running indefinitely
• IT for deterministic final status after outer poll exhaustion and after outer timeout exhaustion