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/services/gateway/admin.go

@@ -58,6 +58,10 @@ type AdminLlmProvider struct {
 	Multimodal bool   `json:"multimodal"`
 	Enabled    bool   `json:"enabled"`
 	Fallback   string `json:"fallback,omitempty"`
+	// MaxTokens is the group's own token budget (0 = unset; the runtime then
+	// falls back to LLM_FALLBACK_MAX_TOKENS on the last fallback step, then the
+	// adapter default).
+	MaxTokens int `json:"maxTokens,omitempty"`
 }
 
 // AdminLlmConfig is the wire shape of GET /api/v1/admin/llm.
@@ -360,6 +364,7 @@ func (s *Server) adminGetLlm(w http.ResponseWriter, r *http.Request) {
 			Multimodal: g.Multimodal,
 			Enabled:    i == 0,
 			Fallback:   g.Fallback,
+			MaxTokens:  g.MaxTokens,
 		})
 	}
 	writeJSON(w, http.StatusOK, AdminLlmConfig{Providers: providers})