feat: add native Gemini support to openai_compat provider

[Ajinkya-Kardile]

Description

This PR enables native integration with Google Gemini by updating the wire_model_for_base_url logic in the openai_compat provider.

Previously, when using a custom OPENAI_BASE_URL (such as the Google Generative Language API endpoint), claw-code would preserve the openai/ prefix in the outgoing JSON payload. This caused API requests to fail because the Google API does not recognize model identifiers like:

models/openai/gemini-....

This change adds gemini to the stripping logic, ensuring that model slugs are correctly normalized to the format expected by the Google Gemini API while maintaining compatibility for other OpenAI-compatible gateways.

---

Commands Run

To verify this change, the following commands were executed inside the rust/ workspace:

cargo fmt --all --check

✅ Passed

cargo clippy --workspace

✅ Passed

cargo test --workspace

✅ Passed

---

Manual Verification

Environment

export OPENAI_BASE_URL="https://generativelanguage.googleapis.com/v1beta/openai/"

Command Executed

./target/release/claw --model openai/gemini-2.5-flash prompt "reply with ready"

Result

✅ Successfully received a response from Gemini.

---

Compatibility Risks

CLI / Provider Behavior

This changes how openai/ model slugs are sanitized when a non-default OPENAI_BASE_URL is detected.

The logic specifically targets models containing "gemini" to avoid prefix corruption while preserving compatibility with existing OpenAI-compatible providers.

Windows / PowerShell

No changes were made to build scripts, shell integrations, or documentation.

The modification is isolated entirely within the Rust api crate.

---

Known Gaps

None.

This implementation follows the existing normalization pattern already used for providers such as:

• xai
• qwen
• kimi

and adheres to the current crate boundaries and provider architecture.