fix(model): 解析 stop_reason + max_tokens 截断重试/非空兜底 — 不再静默空回复

- model.go: Response.StopReason; anthropic wire加stop_reason; openai finish_reason映射(两协议统一,原值透传)
- truncation.go(新,159行): effectiveMaxTokens/retryMaxTokens(×2封顶32768)/isTruncatedEmpty(空正文+无tool_use+stop_reason∈{max_tokens,length})/withEmptyBodyHint/callWithTruncationRetry/callStreamWithTruncationRetry/dropThinking/emitTextSegment
- anthropic.go Chat/ChatStream 走重试+兜底; anthropic_encode.go(新)拆编码(anthropic.go 529→357合规); anthropic_stream message_delta落StopReason
- agent/loop: 把模型层StopReason带到llm.call span的model_stop_reason属性(修掉trace硬编码stop的误导); loop.Result.StopReason语义未动
- 触发仅当截断空正文; tool_use空正文永不触发(native_toolcall_test仍过); 至多1次重试; 兜底回'（思考过长，未能生成回复）'绝不用thinking
- 新增7测试(非流式/流式重试+兜底+tool_use不触发+封顶+openai映射), build/vet/test绿+3平台交叉编译

internal/runtime/model/openai.go

@@ -488,6 +488,7 @@ func responseFromWire(wire *openAIChatResponse) Response {
 		text := wire.Choices[0].Message.Content
 		out.Content = text
 		out.Message = Message{Role: RoleAssistant, Content: []Content{{Type: ContentTypeText, Text: text}}}
+		out.StopReason = wire.Choices[0].FinishReason
 		for _, tc := range wire.Choices[0].Message.ToolCalls {
 			args := map[string]any{}
 			if tc.Function.Arguments != "" {
@@ -546,6 +547,9 @@ func scanSSE(r io.Reader, onEvent func(StreamEvent)) (Response, error) {
 			agg.CompletionTokens = u.CompletionTokens
 			agg.TotalTokens = u.TotalTokens
 		}
+		if len(chunk.Choices) > 0 && chunk.Choices[0].FinishReason != "" {
+			agg.StopReason = chunk.Choices[0].FinishReason
+		}
 		if len(chunk.Choices) == 0 || chunk.Choices[0].Delta == nil {
 			continue
 		}