feat(tokens): 新增 llm_usage 计量账本，按模型统计 token + 仪表盘新旧版切换

- llmrec: 拆分轻量计量与重量 body 录制——token+model+task+worker 永远写入
  新表 llm_usage（不含 body，不受 llm_record 开关控制），body 仍受开关控制
- 修复 token 累加 bug：usage 从覆盖改为 Add（Anthropic 把 input/output 拆到
  message_start/message_delta 两个事件，覆盖会把 input 冲成 0；OpenAI 不受影响）
- 任务总览新增「LLM Token 用量」卡片：按模型统计 + 缓存命中率(缓存读/输入)
- 仪表盘 Token 卡片加「新版/旧版」开关：新版走 llm_usage（逐次精确、含中断消耗、
  真实每日消耗），旧版沿用 activity 统计
- llm_usage 随任务删除保留（历史计量账本）

Claude-Session: https://claude.ai/code/session_014TfZWeVGhjXqZD3vcj44mz

server/commands.go

@@ -4,6 +4,8 @@ import (
 	"net/http"
 	"strconv"
 	"strings"
+
+	"github.com/Autumn-27/artex/db"
 )
 
 // pgListCommands returns tool executions (any tool) from the activity table.
@@ -45,6 +47,53 @@ func (s *Server) pgListLLMRecords(w http.ResponseWriter, r *http.Request) {
 	writeJSON(w, 200, map[string]any{"records": records, "total": total})
 }
 
+// pgTokenByModel returns a task's LLM token usage grouped by model, from the
+// always-on llm_usage metering ledger. Accurate even with per-agent model bindings,
+// pool rotation/failover, and interrupted runs (every call is metered on success or
+// error). No recording toggle required.
+func (s *Server) pgTokenByModel(w http.ResponseWriter, r *http.Request) {
+	id := strings.TrimSpace(r.URL.Query().Get("task"))
+	if id == "" {
+		writeErr(w, 400, "missing task")
+		return
+	}
+	pg := s.m.PG()
+	if pg == nil {
+		writeJSON(w, 200, map[string]any{"models": []db.ModelTokenStat{}})
+		return
+	}
+	models, err := pg.TokenByModel(id)
+	if err != nil {
+		writeErr(w, 500, err.Error())
+		return
+	}
+	writeJSON(w, 200, map[string]any{"models": models})
+}
+
+// pgUsageStats returns global llm_usage aggregates for the dashboard's "new" token
+// view: per-profile totals (all-time) + per-(profile, day) buckets for the daily
+// chart. Sourced from the always-on metering ledger, so it is accurate across
+// per-agent bindings, pool rotation, and interrupted runs.
+func (s *Server) pgUsageStats(w http.ResponseWriter, r *http.Request) {
+	pg := s.m.PG()
+	if pg == nil {
+		writeJSON(w, 200, map[string]any{"by_profile": []db.ProfileUsage{}, "daily": []db.ProfileDayUsage{}})
+		return
+	}
+	days := atoiDefault(r.URL.Query().Get("days"), 365)
+	byProfile, err := pg.UsageByProfile()
+	if err != nil {
+		writeErr(w, 500, err.Error())
+		return
+	}
+	daily, err := pg.UsageDaily(days)
+	if err != nil {
+		writeErr(w, 500, err.Error())
+		return
+	}
+	writeJSON(w, 200, map[string]any{"by_profile": byProfile, "daily": daily})
+}
+
 // pgLLMTasks returns distinct recorded tasks with counts, for the page's task
 // picker (pick a task → filter the list, then delete its conversations).
 func (s *Server) pgLLMTasks(w http.ResponseWriter, r *http.Request) {

server/manager.go

@@ -250,6 +250,9 @@ func NewManager(dir, proxyAddr string) (*Manager, error) {
 	if err := pg.EnsureLLMRecordsTable(); err != nil {
 		log.Printf("[llmrec] create table: %v", err)
 	}
+	if err := pg.EnsureLLMUsageTable(); err != nil {
+		log.Printf("[llmusage] create table: %v", err)
+	}
 	m := &Manager{dir: dir, pg: pg, assets: pg.Assets(), tasks: map[string]*Task{}, interceptor: intercept.New(pg)}
 	if proxyAddr != "" {
 		tr, err := traffic.Open(filepath.Join(dir, "traffic"), proxyAddr)

server/server.go

@@ -690,6 +690,7 @@ func (s *Server) Handler() http.Handler {
 	mux.HandleFunc("GET /api/exploration/tokens", s.tokenStats)
 	mux.HandleFunc("GET /api/tokens/daily", s.tokenDailyStats)
 	mux.HandleFunc("GET /api/tokens/conversations", s.conversationTokens)
+	mux.HandleFunc("GET /api/tokens/usage", s.pgUsageStats) // 全局 llm_usage 聚合（仪表盘新版视图）
 
 	mux.HandleFunc("GET /api/audit", s.getAudit)
 	mux.HandleFunc("POST /api/gc", s.gc)
@@ -702,6 +703,7 @@ func (s *Server) Handler() http.Handler {
 	mux.HandleFunc("GET /api/llm/records", s.pgListLLMRecords)
 	mux.HandleFunc("DELETE /api/llm/records", s.pgDeleteLLMRecords)
 	mux.HandleFunc("GET /api/llm/records/tasks", s.pgLLMTasks)
+	mux.HandleFunc("GET /api/llm/records/by-model", s.pgTokenByModel) // 按模型聚合本任务 token 用量
 	mux.HandleFunc("GET /api/llm/records/{id}", s.pgGetLLMRecord)
 	mux.HandleFunc("GET /api/settings", s.getSettings)
 	mux.HandleFunc("PUT /api/settings", s.putSettings)