fix(sourcehunt): synthesize before stall termination

[ropoctl]

Summary

• detect the no-progress boundary before returning stalled
• deliver one explicit final synthesis/recovery turn at that boundary
• preserve the existing bounded stall terminal if the recovery turn remains unproductive

Rationale

The final-synthesis guard already covered budget and maximum-step boundaries, but a hunter could reach max_steps_without_progress earlier and terminate without ever receiving the intended synthesis instruction. This keeps the existing bounded behavior while ensuring the model gets one chance to report a source-backed lead or conclude cleanly.

Test plan

• uv run --extra dev pytest tests/test_deep_agent_loop.py -q (29 passed)
• uv run --extra dev ruff check clearwing/sourcehunt/hunter.py tests/test_deep_agent_loop.py
• uv run --extra dev ruff format --check clearwing/sourcehunt/hunter.py tests/test_deep_agent_loop.py
• git diff --check