fix(model): DeepSeek 推理参数语义反转 + thinking 开关真发（按官方文档核对结果）

依据 DevDocs/design/04 §3.2「官方核对后的修正」9 条 + 本地 DevDocs/reference/deepseek/facts-for-cyrene.md（官方原句/URL）。
- 🔴 reasoning_effort 白名单改为官方集合 [none, low, high, max] + 别名映射(minimal→low, medium→high, xhigh→high, ultra→max)；**删除就近降级**，非法值一律丢弃+WARN。此前 none(关思考) 被降级成 low(开思考) 属语义反转
- 🔴 DeepSeek thinking 开关真发顶层 {"thinking":{"type":"enabled|disabled"}}（官方默认 enabled，不传≠关）；此前只打 WARN 导致 mode:off 静默无效
- DeepSeek 的 /anthropic 端点忽略 budget_tokens → 不再对其抬 max_tokens（会误导预算/日志），改为按官方用 output_config.effort
- 注释纠正：penalty 属接口级 deprecated（非仅思考模式）；top_p 仅思考模式生效且下限 0.95（不加该字段）
- tool_choice 雷登记：思考模式下 required/指定 tool 会 400（思考默认开）
- 旧模型名清理（deepseek-chat/reasoner 已于 2026-07-24 停用）
- 请求体级测试 + 变异验证（把 none 改回 low / 把 thinking 改回只 WARN → 对应断言 FAIL）

internal/runtime/model/openai_encode.go

@@ -15,12 +15,24 @@ type openAIChatRequest struct {
 	Stream        bool              `json:"stream,omitempty"`
 	StreamOptions *openAIStreamOpts `json:"stream_options,omitempty"`
 	Tools         []openAITool      `json:"tools,omitempty"`
+	// Thinking is DeepSeek's top-level thinking switch. It is sent only for
+	// DeepSeek and only from an explicit switch (or an explicit off that must be
+	// honoured): leaving it out keeps the provider default, which is ENABLED, so
+	// an omitted field is not a way to switch thinking off (design/04 §3.2
+	// fact 2).
+	Thinking *openAIThinking `json:"thinking,omitempty"`
 	// ReasoningEffort is the OpenAI-compatible reasoning strength knob. It is set
 	// only after validation against the detected vendor's whitelist (design/04
-	// §3.2); an unsupported value is downgraded or dropped, never forwarded as-is.
+	// §3.2); an unrecognized value is dropped, never turned into a different tier.
 	ReasoningEffort string `json:"reasoning_effort,omitempty"`
 }
 
+// openAIThinking is the wire form of DeepSeek's thinking switch: enabled or
+// disabled. It is only ever sent to a DeepSeek endpoint.
+type openAIThinking struct {
+	Type string `json:"type"`
+}
+
 // openAIStreamOpts asks the provider to include usage in the final stream chunk.
 type openAIStreamOpts struct {
 	IncludeUsage bool `json:"include_usage"`
@@ -71,10 +83,13 @@ func (a *OpenAIChatAdapter) buildRequest(req Request, stream bool) ([]byte, erro
 		return nil, err
 	}
 	// The vendor behind an OpenAI-compatible endpoint decides which
-	// reasoning_effort tiers are legal; classify it from the resolved model (and
-	// the base URL as a fallback) before touching the request (design/04 §3.2).
+	// reasoning_effort tiers are legal and whether it has a thinking switch;
+	// classify it from the resolved model (and the base URL as a fallback) before
+	// touching the request (design/04 §3.2).
 	vendor := reasoningVendorFor(req.Model, a.baseURL)
-	reportUnmappedThinking(a.log, vendor, req.Model, req.Thinking)
+	effort := resolveReasoningEffort(a.log, vendor, req.ReasoningEffort, req.Model)
+	thinking, effort := resolveOpenAIReasoning(a.log, vendor, req.Model, req.Thinking, effort)
+	reportForcedToolChoiceWithThinking(a.log, req.Model, req.ToolChoice, openAIThinkingOn(thinking, effort))
 	wire := openAIChatRequest{
 		Model:           req.Model,
 		Messages:        msgs,
@@ -82,7 +97,8 @@ func (a *OpenAIChatAdapter) buildRequest(req Request, stream bool) ([]byte, erro
 		MaxTokens:       req.MaxTokens,
 		Stream:          stream,
 		Tools:           buildTools(req.Tools),
-		ReasoningEffort: resolveReasoningEffort(a.log, vendor, req.ReasoningEffort, req.Model),
+		Thinking:        thinking,
+		ReasoningEffort: effort,
 	}
 	if stream {
 		wire.StreamOptions = &openAIStreamOpts{IncludeUsage: true}