Add OpenAI Responses API (/v1/responses) endpoint

Translate Responses API requests (input→messages, instructions→system,
max_output_tokens→max_tokens, text.format→response_format, flat tools→nested
{function:{…}}) to upstream chat/completions, then convert responses back to
Responses format (streaming SSE event lifecycle + non-streaming JSON).

Verified against OpenAI migration guide and Python SDK Response model:
- Echo back required fields parallel_tool_calls/tool_choice/tools
- Include content:[] in reasoning items, logprobs:[] in output_text parts
- Support function_call/function_call_output multi-turn input items
- Map usage fields prompt_tokens→input_tokens, completion_tokens→output_tokens

internal/server/server.go

@@ -65,6 +65,7 @@ func (s *Server) routes() {
 	s.mux.HandleFunc("POST /api/models/test", s.withLoginSession(s.testModel))
 	s.mux.HandleFunc("GET /v1/models", s.withAPIKey(s.models))
 	s.mux.HandleFunc("POST /v1/chat/completions", s.withAPIKey(s.chatCompletions))
+	s.mux.HandleFunc("POST /v1/responses", s.withAPIKey(s.responses))
 }
 
 func (s *Server) healthz(w http.ResponseWriter, r *http.Request) {
@@ -704,9 +705,9 @@ func (s *Server) testModel(w http.ResponseWriter, r *http.Request) {
 			payload := sseDataPayload(line)
 			if payload != "" && payload != "[DONE]" {
 				var evt struct {
-					State         string `json:"state"`
-					ErrorMessage  string `json:"errorMessage"`
-					Choices       []struct {
+					State        string `json:"state"`
+					ErrorMessage string `json:"errorMessage"`
+					Choices      []struct {
 						Delta struct {
 							Content string `json:"content"`
 						} `json:"delta"`