feat(relay): hosted-tool conversion fidelity, reasoning normalization, and billing usage integrity (#7137)

* feat(relaykit): preserve hosted tools across conversions

- add protocol-neutral hosted-tool DTOs, conversion metadata, and loss policies
- bridge citations, grounding metadata, and hosted-tool stream lifecycles
- document the public conversion behavior and channel policy controls

* refactor(relaykit): normalize reasoning and thinking intent

- centralize provider-neutral reasoning intent, effort, and budget mappings
- parse model suffixes at the host entry boundary while preserving provider-owned tails
- keep adaptive Claude thinking and explicit zero-token compatibility consistent

* fix(billing): preserve authoritative usage across relay hops

- carry native BillingUsage sidecars through direct and streamed protocol bridges
- merge partial and terminal usage monotonically with safe fallback settlement
- retain cache metadata, penultimate usage, and per-call Gemini tool surcharges

* feat(relay): bridge Responses with Claude and Gemini protocols

- add direct request, response, and stream converters across supported relay formats
- expose Claude count_tokens and Chat-to-Responses compatibility endpoints
- carry conversion diagnostics through the host while retaining the curated public goldens

* fix(relay): wire relaykit conversions into host channels

- connect handlers, adaptors, and channel settings to the standalone conversion layer
- keep model mapping, pricing identity, retries, and provider-specific suffix behavior aligned
- ignore local audit artifacts and retain focused public regression coverage

relaykit/dto/openai_request.go

@@ -81,7 +81,7 @@ type GeneralOpenAIRequest struct {
 	ExtraBody json.RawMessage `json:"extra_body,omitempty"`
 	//xai
 	SearchParameters json.RawMessage `json:"search_parameters,omitempty"`
-	// claude
+	// OpenAI Chat web search.
 	WebSearchOptions *WebSearchOptions `json:"web_search_options,omitempty"`
 	// OpenRouter Params
 	Usage     json.RawMessage `json:"usage,omitempty"`
@@ -108,6 +108,9 @@ type GeneralOpenAIRequest struct {
 	ReasoningSplit json.RawMessage `json:"reasoning_split,omitempty"`
 	// vLLM
 	ThinkingTokenBudget json.RawMessage `json:"thinking_token_budget,omitempty"`
+
+	// Internal conversion state; never serialized to an upstream protocol.
+	ReasoningConversion *ReasoningConversionState `json:"-"`
 }
 
 func (r GeneralOpenAIRequest) MarshalJSON() ([]byte, error) {
@@ -266,6 +269,7 @@ type FunctionRequest struct {
 	Name        string `json:"name"`
 	Parameters  any    `json:"parameters,omitempty"`
 	Arguments   string `json:"arguments,omitempty"`
+	Strict      *bool  `json:"strict,omitempty"`
 }
 
 type StreamOptions struct {
@@ -311,7 +315,10 @@ type Message struct {
 	Reasoning        *string         `json:"reasoning,omitempty"`
 	ToolCalls        json.RawMessage `json:"tool_calls,omitempty"`
 	ToolCallId       string          `json:"tool_call_id,omitempty"`
-	parsedContent    []MediaContent
+	// Annotations is an official Chat response field. Keeping it on the shared
+	// message type also preserves annotations when clients replay assistant output.
+	Annotations   json.RawMessage `json:"annotations,omitempty"`
+	parsedContent []MediaContent
 	//parsedStringContent *string
 }
 
@@ -485,14 +492,14 @@ func (m *Message) ParseToolCalls() []ToolCallRequest {
 		return nil
 	}
 	var toolCalls []ToolCallRequest
-	if err := json.Unmarshal(m.ToolCalls, &toolCalls); err == nil {
+	if err := kitutil.Unmarshal(m.ToolCalls, &toolCalls); err == nil {
 		return toolCalls
 	}
 	return toolCalls
 }
 
 func (m *Message) SetToolCalls(toolCalls any) {
-	toolCallsJson, _ := json.Marshal(toolCalls)
+	toolCallsJson, _ := kitutil.Marshal(toolCalls)
 	m.ToolCalls = toolCallsJson
 }
 
@@ -562,6 +569,11 @@ func (m *Message) ParseContent() []MediaContent {
 		return contentList
 	}
 
+	if content, ok := m.Content.([]MediaContent); ok {
+		m.parsedContent = content
+		return content
+	}
+
 	// 尝试解析为数组
 	//var arrayContent []map[string]interface{}
 
@@ -682,7 +694,7 @@ func (m *Message) ParseContent() []MediaContent {
 	}
 
 	var stringContent string
-	if err := json.Unmarshal(m.Content, &stringContent); err == nil {
+	if err := kitutil.Unmarshal(m.Content, &stringContent); err == nil {
 		m.parsedStringContent = &stringContent
 		return stringContent
 	}
@@ -707,14 +719,14 @@ func (m *Message) SetNullContent() {
 }
 
 func (m *Message) SetStringContent(content string) {
-	jsonContent, _ := json.Marshal(content)
+	jsonContent, _ := kitutil.Marshal(content)
 	m.Content = jsonContent
 	m.parsedStringContent = &content
 	m.parsedContent = nil
 }
 
 func (m *Message) SetMediaContent(content []MediaContent) {
-	jsonContent, _ := json.Marshal(content)
+	jsonContent, _ := kitutil.Marshal(content)
 	m.Content = jsonContent
 	m.parsedContent = nil
 	m.parsedStringContent = nil
@@ -725,7 +737,7 @@ func (m *Message) IsStringContent() bool {
 		return true
 	}
 	var stringContent string
-	if err := json.Unmarshal(m.Content, &stringContent); err == nil {
+	if err := kitutil.Unmarshal(m.Content, &stringContent); err == nil {
 		m.parsedStringContent = &stringContent
 		return true
 	}
@@ -741,7 +753,7 @@ func (m *Message) ParseContent() []MediaContent {
 
 	// 先尝试解析为字符串
 	var stringContent string
-	if err := json.Unmarshal(m.Content, &stringContent); err == nil {
+	if err := kitutil.Unmarshal(m.Content, &stringContent); err == nil {
 		contentList = []MediaContent{{
 			Type: ContentTypeText,
 			Text: stringContent,
@@ -752,7 +764,7 @@ func (m *Message) ParseContent() []MediaContent {
 
 	// 尝试解析为数组
 	var arrayContent []map[string]interface{}
-	if err := json.Unmarshal(m.Content, &arrayContent); err == nil {
+	if err := kitutil.Unmarshal(m.Content, &arrayContent); err == nil {
 		for _, contentItem := range arrayContent {
 			contentType, ok := contentItem["type"].(string)
 			if !ok {
@@ -907,6 +919,9 @@ type OpenAIResponsesRequest struct {
 	ThinkingBudget json.RawMessage `json:"thinking_budget,omitempty"`
 	// perplexity
 	Preset json.RawMessage `json:"preset,omitempty"`
+
+	// Internal conversion state; never serialized to an upstream protocol.
+	ReasoningConversion *ReasoningConversionState `json:"-"`
 }
 
 func (r OpenAIResponsesRequest) MarshalJSON() ([]byte, error) {