feat: Phase 2 - reliability & observability

- Circuit breaker: proper thresholds (3 failures, 30s timeout)
- HTTP timeouts: 30s on all providers (was no timeout)
- Structured logging: slog replaces fmt.Printf throughout
- Stream errors: propagated as SSE error events to client
- Registry fetch: retry with backoff (3 attempts)
- Registry reads in dashboard protected by RWMutex

internal/models/models.go

@@ -26,12 +26,12 @@ type ChatCompletionRequest struct {
 }
 
 type ChatMessage struct {
-	Role             string         `json:"role"` // "system", "user", "assistant", "tool"
-	Content          interface{}    `json:"content"`
-	ReasoningContent *string        `json:"reasoning_content,omitempty"`
-	ToolCalls        []ToolCall     `json:"tool_calls,omitempty"`
-	Name             *string        `json:"name,omitempty"`
-	ToolCallID       *string        `json:"tool_call_id,omitempty"`
+	Role             string      `json:"role"` // "system", "user", "assistant", "tool"
+	Content          interface{} `json:"content"`
+	ReasoningContent *string     `json:"reasoning_content,omitempty"`
+	ToolCalls        []ToolCall  `json:"tool_calls,omitempty"`
+	Name             *string     `json:"name,omitempty"`
+	ToolCallID       *string     `json:"tool_call_id,omitempty"`
 }
 
 type ContentPart struct {
@@ -53,9 +53,9 @@ type Tool struct {
 }
 
 type FunctionDef struct {
-	Name       string          `json:"name"`
+	Name        string          `json:"name"`
 	Description *string         `json:"description,omitempty"`
-	Parameters json.RawMessage `json:"parameters,omitempty"`
+	Parameters  json.RawMessage `json:"parameters,omitempty"`
 }
 
 type ToolCall struct {
@@ -116,6 +116,7 @@ type ChatCompletionStreamResponse struct {
 	Model   string             `json:"model"`
 	Choices []ChatStreamChoice `json:"choices"`
 	Usage   *Usage             `json:"usage,omitempty"`
+	Error   *string            `json:"error,omitempty"`
 }
 
 type ChatStreamChoice struct {