fix(v1): retain Terminus reasoning history

[parkerpettit]

Summary

• Enable Harbor interleaved thinking for the Terminus-2 adapter.
• Preserve reasoning content in later prompts so unchanged multi-turn histories remain one training branch.

Test

• uv run ruff check verifiers/v1/harnesses/terminus_2/program.py
• Confirmed with Harbor 0.21.0 that the second request contains the first response reasoning content.

Note

Pass interleaved_thinking=True to Terminus2 agent in main

Adds the interleaved_thinking=True keyword argument when constructing the Terminus2 instance in program.py. This retains the agent's reasoning history across interleaved thinking steps.

^{Macroscope summarized e22fc88.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This one-line change enables reasoning content to persist into later Terminus-2 model requests for every rollout, altering existing conversation history and downstream model inputs. Because it activates non-trivial behavior on an existing path rather than an opt-in path, human review is appropriate.

^{You can add or adjust custom eligibility rules. Learn more.}