feat(providers): model-registry routing + Responses API support and streaming fallbacks for OpenAI/Gemini

src/utils/registry.rs

@@ -22,3 +22,28 @@ pub async fn fetch_registry() -> Result<ModelRegistry> {
 
     Ok(registry)
 }
+
+/// Heuristic: decide whether a model should be routed to OpenAI Responses API
+/// instead of the legacy chat/completions endpoint.
+///
+/// Currently this uses simple patterns (codex, gpt-5 series) and also checks
+/// the loaded registry metadata name for the substring "codex" as a hint.
+pub fn model_prefers_responses(registry: &ModelRegistry, model: &str) -> bool {
+    let model_lc = model.to_lowercase();
+
+    if model_lc.contains("codex") {
+        return true;
+    }
+
+    if model_lc.starts_with("gpt-5") || model_lc.contains("gpt-5.") {
+        return true;
+    }
+
+    if let Some(meta) = registry.find_model(model) {
+        if meta.name.to_lowercase().contains("codex") {
+            return true;
+        }
+    }
+
+    false
+}