Subscription-based model graceful fallback on token limit being reached

Problem

When a user is using Claude subscription mode, a run on even the simplest smart contract project results in a their 5-hour limit being reached. We don't have a graceful way to handle token limit errors. Typically, a resume command needs to be issued once the user understands what went wrong in their run

Proposed Solution

1. Set a threshold for token usage (requires Ultrafuzz to ping model while a node is running)
2. Pause the node when the token threshold is reached.
3. Provide details to user when inspecting a run to understand why the node paused.

Anything Else We Should Know

For now I have a proposed solution I use locally. I know a lot of activity is going on in the project and I don't consider this a high-priority task. That said, if you would like for me to push a PR sooner than later let me know. Or if this is not something needed in the project I can close this ticket for now.

The issue names no files or tests; start by tracing the node execution flow and the run-inspection entry point. Determine how token usage and subscription-limit failures are surfaced, then verify that a threshold pauses the node and that run details explain the pause before resume can continue it.