fix(relay): preserve presence/frequency penalty in Responses conversion (#6654)

relaykit/dto/openai_request.go

@@ -867,6 +867,11 @@ type OpenAIResponsesRequest struct {
 	Metadata           json.RawMessage `json:"metadata,omitempty"`
 	Moderation         json.RawMessage `json:"moderation,omitempty"`
 	ParallelToolCalls  json.RawMessage `json:"parallel_tool_calls,omitempty"`
+	// FrequencyPenalty/PresencePenalty are not part of the official OpenAI
+	// Responses API; they are forwarded verbatim for OpenAI-compatible upstreams
+	// (e.g. vLLM) that accept them.
+	FrequencyPenalty   json.RawMessage `json:"frequency_penalty,omitempty"`
+	PresencePenalty    json.RawMessage `json:"presence_penalty,omitempty"`
 	PreviousResponseID string          `json:"previous_response_id,omitempty"`
 	Reasoning          *Reasoning      `json:"reasoning,omitempty"`
 	// ServiceTier specifies upstream service level and may affect billing.

relaykit/dto/openai_request_zero_value_test.go

@@ -123,7 +123,9 @@ func TestOpenAIResponsesRequestPreserveExplicitZeroValues(t *testing.T) {
 		"max_output_tokens":0,
 		"max_tool_calls":0,
 		"stream":false,
-		"top_p":0
+		"top_p":0,
+		"frequency_penalty":0,
+		"presence_penalty":0
 	}`)
 
 	var req OpenAIResponsesRequest
@@ -137,6 +139,8 @@ func TestOpenAIResponsesRequestPreserveExplicitZeroValues(t *testing.T) {
 	require.True(t, gjson.GetBytes(encoded, "max_tool_calls").Exists())
 	require.True(t, gjson.GetBytes(encoded, "stream").Exists())
 	require.True(t, gjson.GetBytes(encoded, "top_p").Exists())
+	require.True(t, gjson.GetBytes(encoded, "frequency_penalty").Exists())
+	require.True(t, gjson.GetBytes(encoded, "presence_penalty").Exists())
 }
 
 func TestOpenAIResponsesRequestPreserveQwenThinkingBudget(t *testing.T) {