fix(opencode-zen): route GPT-5.6 models through Responses API

[prabhtheone]

what does this pr fixes

This PR fixes GPT-5.6 model routing for the OpenCode Zen provider.

The GPT-5.6 models are served through the OpenAI Responses API, but the OpenCode Zen registry did not explicitly declare their targetFormat. As a result, OmniRoute could fall back to the Chat Completions endpoint instead of /responses.

Models affected

• gpt-5.6-sol
• gpt-5.6-terra
• gpt-5.6-luna

Changes

• Added targetFormat: "openai-responses" to all three GPT-5.6 models.
• Added a regression test covering all GPT-5.6 model entries.
• Verified that resolveOpencodeTargetFormat() resolves the models to the Responses API through the opencode-zen provider alias.

Why?

Without the correct target format, requests can be sent to the wrong upstream endpoint, causing GPT-5.6 requests to fail instead of being routed through the supported Responses API.

Testing

Added:

tests/unit/opencode-zen-gpt56-responses-14229.test.ts

The test verifies that all three GPT-5.6 models:

1. Declare targetFormat: "openai-responses".
2. Resolve to openai-responses through the OpenCode Zen provider.

Related issue

Fixes #14229

[diegosouzapw]

Thank you for tracking down the same #14229 bug — your diff is correct and matches what we're shipping. There's an earlier PR (#14230) open for the exact same fix that we're merging instead (it also adds a changelog entry and a control test), so I'm closing this one as a duplicate rather than merging both. Appreciate the report and the regression test either way — feel free to route future opencode-zen catches our way.