feat(model): LLM token 预算可配置化 — models.json每模型maxTokens + LLM_FALLBACK_MAX_TOKENS回退兜底

- config.go: LLMConfig.FallbackMaxTokens(LLM_FALLBACK_MAX_TOKENS, 默认8192)
- runtimeconfig/models.go: Group.MaxTokens(models.json每模型独立预算, 负值拒绝)
- model/fallback.go: Chain.WithTokenBudget + applyBudget; runtime.go buildSnapshot接入(active.MaxTokens, env.FallbackMaxTokens)
- 优先级: 显式Request.MaxTokens > 组maxTokens > 仅多步链末端的LLM_FALLBACK_MAX_TOKENS > 适配器默认8192; 单步链不动env兜底(避免OpenAI主模型行为回归)
- gateway admin: AdminLlmProvider.MaxTokens 透传/回显/负值400
- .env.example + .env(LLM_FALLBACK_MAX_TOKENS=8192); Docs/api-client v0.27 + api-reference + design/04
- 新增测试(config/runtimeconfig/budget_test/admin_write), build/vet/test绿+3平台交叉编译

internal/config/runtimeconfig/models.go

@@ -42,6 +42,11 @@ type Group struct {
 	Model      string `json:"model"`
 	Fallback   string `json:"fallback,omitempty"`
 	Multimodal bool   `json:"multimodal,omitempty"`
+	// MaxTokens is the group's own token budget, applied to every model in the
+	// group's fallback chain (primary and fallback). Zero means "unset": the
+	// last-resort fallback step then falls back to LLM_FALLBACK_MAX_TOKENS, and
+	// beyond that the adapter default applies.
+	MaxTokens int `json:"maxTokens,omitempty"`
 }
 
 // ErrInvalidGroup is returned when a group fails schema validation.
@@ -67,6 +72,9 @@ func (g Group) Validate() error {
 	if strings.TrimSpace(g.Model) == "" {
 		return fmt.Errorf("model is required")
 	}
+	if g.MaxTokens < 0 {
+		return fmt.Errorf("maxTokens must be >= 0, got %d", g.MaxTokens)
+	}
 	return nil
 }