fix(dialect): 修正 OpenAI 流式用量终态

internal/dialect/openai_usage.go

@@ -142,6 +142,22 @@ func openAIUsageFinal(root map[string]json.RawMessage) bool {
 	if !exists {
 		return false
 	}
-	var choices []json.RawMessage
-	return json.Unmarshal(rawChoices, &choices) == nil && choices != nil && len(choices) == 0
+	var choices []map[string]json.RawMessage
+	if json.Unmarshal(rawChoices, &choices) != nil || choices == nil {
+		return false
+	}
+	if len(choices) == 0 {
+		return true
+	}
+	for _, choice := range choices {
+		rawReason, exists := choice["finish_reason"]
+		if !exists {
+			continue
+		}
+		var reason string
+		if json.Unmarshal(rawReason, &reason) == nil && reason != "" {
+			return true
+		}
+	}
+	return false
 }

internal/dialect/openai_usage_test.go

@@ -313,9 +313,18 @@ func TestUsageOpenAIStreamSnapshotsAndFinality(t *testing.T) {
 			state: usage.StateComplete,
 			want:  usage.Tokens{UncachedInput: 100, Output: 30},
 		},
+		{
+			name: "terminal choice with usage is complete",
+			steps: []string{
+				`{"choices":[{"index":0,"delta":{"content":"你好"},"finish_reason":null}]}`,
+				`{"choices":[{"index":0,"delta":{},"finish_reason":"stop","native_finish_reason":"stop"}],"usage":{"completion_tokens":13,"total_tokens":317,"prompt_tokens":304,"prompt_tokens_details":{"cached_tokens":0,"cached_creation_tokens":0},"completion_tokens_details":{"reasoning_tokens":0}}}`,
+			},
+			state: usage.StateComplete,
+			want:  usage.Tokens{UncachedInput: 304, Output: 13},
+		},
 		{
 			name:  "usage without final marker is partial",
-			steps: []string{`{"choices":[{"delta":{}}],"usage":{"prompt_tokens":100,"completion_tokens":30}}`},
+			steps: []string{`{"choices":[{"delta":{},"finish_reason":null}],"usage":{"prompt_tokens":100,"completion_tokens":30}}`},
 			state: usage.StatePartial,
 			want:  usage.Tokens{UncachedInput: 100, Output: 30},
 		},

internal/gateway/usage_capture_test.go

@@ -398,6 +398,15 @@ func TestForwardStreamUsageStateRequiresProviderFinalEvidence(t *testing.T) {
 		want    usage.State
 	}{
 		{name: "OpenAI partial", dialect: dialect.NewOpenAI(http.DefaultClient), path: "/v1/chat/completions", wire: `data: {"choices":[{"delta":{}}],"usage":{"prompt_tokens":100,"completion_tokens":30,"prompt_tokens_details":{"cached_tokens":20}}}` + "\n\n", want: usage.StatePartial},
+		{
+			name:    "OpenAI terminal choice",
+			dialect: dialect.NewOpenAI(http.DefaultClient),
+			path:    "/v1/chat/completions",
+			wire: "data: {\"choices\":[{\"index\":0,\"delta\":{\"content\":\"你好\"},\"finish_reason\":null}]}\n\n" +
+				"data: {\"choices\":[{\"index\":0,\"delta\":{},\"finish_reason\":\"stop\",\"native_finish_reason\":\"stop\"}],\"usage\":{\"completion_tokens\":13,\"total_tokens\":317,\"prompt_tokens\":304,\"prompt_tokens_details\":{\"cached_tokens\":0,\"cached_creation_tokens\":0},\"completion_tokens_details\":{\"reasoning_tokens\":0}}}\n\n" +
+				"data: [DONE]\n\n",
+			want: usage.StateComplete,
+		},
 		{name: "Anthropic partial", dialect: dialect.NewAnthropic(http.DefaultClient), path: "/v1/messages", wire: `data: {"type":"message_start","message":{"usage":{"input_tokens":80,"cache_read_input_tokens":20}}}` + "\n\n", want: usage.StatePartial},
 		{name: "Gemini partial", dialect: dialect.NewGemini(http.DefaultClient), path: "/v1beta/models/gemini:streamGenerateContent", wire: `data: {"usageMetadata":{"promptTokenCount":100,"cachedContentTokenCount":20,"candidatesTokenCount":30}}` + "\n\n", want: usage.StatePartial},
 		{name: "missing", dialect: dialect.NewOpenAI(http.DefaultClient), path: "/v1/chat/completions", wire: "data: {\"choices\":[{\"delta\":{}}]}\n\n", want: usage.StateMissing},