fix(kiro): map GPT reasoning effort fields

[SemonCat]

Summary

• map GPT-5.6 effort to additionalModelRequestFields.reasoning.effort
• preserve Claude adaptive-effort mapping through output_config.effort
• keep the existing prompt tags for backward compatibility

Why

Kiro CLI selects the effort field path from each model schema. GPT models use reasoning.effort, while Claude models use output_config.effort. 9Router previously only emitted the Claude-shaped request fields, so GPT-5.6 did not receive the same structured thinking level as Kiro CLI.

Verification

• production Docker build (npm run build) passes on Node 22
• targeted mapping smoke confirms GPT reasoning.effort and existing Claude output_config.effort
• live Kiro GPT-5.6 request with reasoning_effort: high returned HTTP 200
• added OpenAI- and Claude-format translator coverage for GPT-5.6 effort mapping and invalid values

[SemonCat]

Superseded by #2707, which also preserves Responses API reasoning.effort through the OpenAI bridge and covers the actual Hermes codex_responses path.

[decolua]

Thanks @SemonCat for the contribution! Reviewed and merged into master. 🙏