fix(agent): add holding pattern supervisor detection to prevent conversational stalls

[Katsoragi]

Summary

Adds holding pattern supervisor detection to the agent loop to prevent conversational stalls. It detects when the model writes "holding" phrases (e.g., "one moment", "please wait", "I am now reading/searching/fetching...") but ends the turn without executing any tools. It intercepts this and injects a system nudge instructing the model to output the tool call directly, preventing infinite conversational loops.

Target branch

• This PR targets dev, not main. All PRs land in dev; main is curated by the maintainer at each release. If your PR is on main by accident, click "Edit" on this PR and change the base.

Linked Issue

Fixes #5162

Type of Change

• Bug fix (non-breaking — fixes a confirmed issue)
• New feature (non-breaking — adds new behaviour)

Checklist

• I searched open issues and open PRs — this is not a duplicate.
• This PR targets dev
• My changes are limited to the scope described above — no unrelated refactors or whitespace changes mixed in.
• I actually ran the app and verified the change works end-to-end.

How to Test

Verify that if the model output contains a holding pattern without a tool call, the supervisor successfully injects a nudge and continues.

Checks Run

• Run pytest suite: pytest tests/test_agent_loop.py tests/test_loop_breaker_runaway.py (All Passed).