fix(openai): emit valid responses stream error chunks

When /v1/responses streaming fails after headers are sent, we now emit a type=error chunk instead of an HTTP-style {error:{...}} payload, preventing AI SDK chunk validation errors.

sdk/api/handlers/openai_responses_stream_error_test.go

@@ -0,0 +1,48 @@
+package handlers
+
+import (
+	"encoding/json"
+	"net/http"
+	"testing"
+)
+
+func TestBuildOpenAIResponsesStreamErrorChunk(t *testing.T) {
+	chunk := BuildOpenAIResponsesStreamErrorChunk(http.StatusInternalServerError, "unexpected EOF", 0)
+	var payload map[string]any
+	if err := json.Unmarshal(chunk, &payload); err != nil {
+		t.Fatalf("unmarshal: %v", err)
+	}
+	if payload["type"] != "error" {
+		t.Fatalf("type = %v, want %q", payload["type"], "error")
+	}
+	if payload["code"] != "internal_server_error" {
+		t.Fatalf("code = %v, want %q", payload["code"], "internal_server_error")
+	}
+	if payload["message"] != "unexpected EOF" {
+		t.Fatalf("message = %v, want %q", payload["message"], "unexpected EOF")
+	}
+	if payload["sequence_number"] != float64(0) {
+		t.Fatalf("sequence_number = %v, want %v", payload["sequence_number"], 0)
+	}
+}
+
+func TestBuildOpenAIResponsesStreamErrorChunkExtractsHTTPErrorBody(t *testing.T) {
+	chunk := BuildOpenAIResponsesStreamErrorChunk(
+		http.StatusInternalServerError,
+		`{"error":{"message":"oops","type":"server_error","code":"internal_server_error"}}`,
+		0,
+	)
+	var payload map[string]any
+	if err := json.Unmarshal(chunk, &payload); err != nil {
+		t.Fatalf("unmarshal: %v", err)
+	}
+	if payload["type"] != "error" {
+		t.Fatalf("type = %v, want %q", payload["type"], "error")
+	}
+	if payload["code"] != "internal_server_error" {
+		t.Fatalf("code = %v, want %q", payload["code"], "internal_server_error")
+	}
+	if payload["message"] != "oops" {
+		t.Fatalf("message = %v, want %q", payload["message"], "oops")
+	}
+}