fix(acp): align reasoning effort capabilities

[szysammie]

Fixes #945

Summary

• Upgrade Codex ACP, Claude ACP, Gemini CLI, Pi ACP, OpenClaw, OpenCode, and MiMo CLI pins.
• Dispatch reasoning effort through live ACP capabilities: reasoning_effort (Codex), effort (Claude), and thought_level (Pi).
• Configure launch-time effort for OpenHands and Harvey LAB; leave agents without a confirmed effort interface fail-closed.

Root cause

BenchFlow’s prior Codex ACP pin and runtime assumed the legacy model[effort] protocol. Current Codex ACP exposes independent model and reasoning_effort options, so effort could not be correctly configured. The same review found launch-configured agents whose declared effort setting was not injected before process startup.

Companion manifest PR

• chore(acp): sync agent runtime version pins agents#60 synchronizes the manifest-owned runtime version pins required by the manifest-parity gate.
• The parity check will become green after that PR is merged into benchflow-ai/agents@main and this PR’s workflow is rerun.

Validation

• Original regression suite: 302 passed, 2 skipped.
• Focused ACP, registry, environment, and live manifest parity suite: 70 passed.
• Agents contract suite: 21 passed.
• ruff check src tests tools.
• ruff format --check src tests tools.
• ty check src.

[szysammie]

Hi! Could a maintainer please approve the pending GitHub Actions workflows for this external contribution? Thank you.

[xdotli]

sure!

[szysammie]

Thanks for reviewing this PR. Because BenchFlow’s manifest-parity check depends on benchflow-ai/agents@main, I have opened the companion manifest update in benchflow-ai/agents#60.

Could a maintainer please review and approve the workflow runs for that PR first? If its checks pass, please merge it so the updated manifests are available on agents/main. Afterward, please approve and re-run the checks on this PR so #946 can validate against the merged agent manifests. Thank you!

[xdotli]

approved!

Thanks for reviewing this PR. Because BenchFlow’s manifest-parity check depends on benchflow-ai/agents@main, I have opened the companion manifest update in benchflow-ai/agents#60.

Could a maintainer please review and approve the workflow runs for that PR first? If its checks pass, please merge it so the updated manifests are available on agents/main. Afterward, please approve and re-run the checks on this PR so #946 can validate against the merged agent manifests. Thank you!

[xdotli]

and thanks for the contribution