feat(activity): 丢弃记录日志合并为单条 error,带 reason/summary/累计数

- 把「丢弃该记录」与「FK 诊断」合并成一条 error 级日志(含"丢弃"关键字,
  后台只筛 error 也能看到完整信息,不再漏掉 info 级诊断行)
- reason 分类:fk_violation(23503) / unique_violation(23505) / pg_error(code) / write_error
- 带 summary 预览(截断去换行)+ 该任务累计丢弃条数,便于判断丢的是哪条、丢弃规模
- 新增 dropReason/bumpDrop/preview 辅助与 dropCnt 计数

server/engine.go

@@ -8,6 +8,7 @@ import (
 	"strconv"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"github.com/Autumn-27/artex/agent"
@@ -25,6 +26,44 @@ func isFKViolation(err error) bool {
 	return errors.As(err, &pgErr) && pgErr.Code == "23503"
 }
 
+// dropReason classifies why an activity write was dropped, so the log can be
+// grouped/analysed by cause rather than by raw error text.
+func dropReason(err error) string {
+	var pgErr *pgconn.PgError
+	if errors.As(err, &pgErr) {
+		switch pgErr.Code {
+		case "23503":
+			return "fk_violation(23503,父exploration不存在)"
+		case "23505":
+			return "unique_violation(23505)"
+		default:
+			return "pg_error(" + pgErr.Code + ")"
+		}
+	}
+	return "write_error"
+}
+
+// bumpDrop increments and returns the running count of dropped (unpersistable)
+// activity records for a task. Concurrent planner + worker emits race here, so the
+// counter is an atomic behind sync.Map. The count in the log shows loss scale at a
+// glance instead of forcing a grep-and-count.
+func (e *Engine) bumpDrop(taskID string) int64 {
+	v, _ := e.dropCnt.LoadOrStore(taskID, new(int64))
+	return atomic.AddInt64(v.(*int64), 1)
+}
+
+// preview collapses newlines and trims s to a short rune-safe snippet for one-line
+// log output (avoids dumping a multi-KB summary/detail into the log).
+func preview(s string, n int) string {
+	s = strings.ReplaceAll(s, "\r", " ")
+	s = strings.ReplaceAll(s, "\n", " ")
+	r := []rune(s)
+	if len(r) > n {
+		return string(r[:n]) + "…"
+	}
+	return string(r)
+}
+
 // model_error（provider/API 故障：LLM 层瞬时重试耗尽，或流已开始后中途断流）
 // 收场的 work 不是「试过没做完」也不是真失败，而是外部抖动。默认把它当永久
 // blocked 会白丢一条意图，所以这里对该终态额外重跑几次，每次之间退避一下，给
@@ -53,6 +92,7 @@ type Engine struct {
 	started sync.Map // taskID -> bool, so Run is idempotent per task
 	lastAct sync.Map // taskID -> int64 unix, last planner/worker activity (heartbeat)
 	paused  sync.Map // taskID -> bool, user-paused (planner + workers idle but loops alive)
+	dropCnt sync.Map // taskID -> *int64, running count of dropped (unpersistable) activity records
 
 	// per-task execution context: each planner.Plan / worker.Execute runs under it,
 	// so pausing can CANCEL an in-flight run (not just skip the next one). Recreated
@@ -279,22 +319,26 @@ func (e *Engine) emitActivity(t *Task, r db.Activity) {
 	id, err := e.appendActivity(t, r)
 	if err != nil {
 		// NO LONGER SILENT: dropping a record breaks command↔result pairing in the
-		// trace — a tool_use whose tool_result was lost shows as "执行中" forever.
-		log.Printf("[activity] task %s 重试后仍失败、丢弃该记录 (kind=%s tool=%s tuid=%s): %v",
-			t.ID, r.Kind, r.Tool, r.ToolUseID, err)
-		// On the FK-parent failure (23503) probe the live DB so the next occurrence
-		// records WHY the exploration is unreachable instead of just that it is.
+		// trace — a tool_use whose tool_result was lost shows as "执行中" forever, and
+		// a lost 'result'/'round' record leaves the session with no summary ("无总结").
+		// Everything needed to分析根因 goes into ONE error-level line: reason class,
+		// summary preview, running drop count for this task, and — on the FK case — a
+		// live probe of WHY the parent exploration is unreachable.
+		n := e.bumpDrop(t.ID)
+		diag := ""
+		// On the FK-parent failure (23503) probe the live DB so the log records WHY the
+		// exploration is unreachable (row gone / wrong expID) instead of just that it is.
 		if isFKViolation(err) {
 			storeID := t.Store.ID()
-			exists, refs, maxID, dErr := e.m.pg.ExplorationDiag(storeID)
-			if dErr != nil {
-				log.Printf("[activity] task %s FK 诊断查询失败 (store.expID=%d task.ExpID=%d): %v",
-					t.ID, storeID, t.ExpID, dErr)
+			if exists, refs, maxID, dErr := e.m.pg.ExplorationDiag(storeID); dErr != nil {
+				diag = fmt.Sprintf(" | FK诊断查询失败(store.expID=%d task.ExpID=%d): %v", storeID, t.ExpID, dErr)
 			} else {
-				log.Printf("[activity] task %s FK 诊断: store.expID=%d task.ExpID=%d exploration存在=%v 引用它的task数=%d 当前MAX(exploration.id)=%d",
-					t.ID, storeID, t.ExpID, exists, refs, maxID)
+				diag = fmt.Sprintf(" | FK诊断: store.expID=%d task.ExpID=%d exploration存在=%v 引用它的task数=%d MAX(exploration.id)=%d",
+					storeID, t.ExpID, exists, refs, maxID)
 			}
 		}
+		log.Printf("[activity] task %s 丢弃活动记录(该任务累计第 %d 条) worker=%s kind=%s tool=%s tuid=%s reason=%s summary=%q: %v%s",
+			t.ID, n, r.Worker, r.Kind, r.Tool, r.ToolUseID, dropReason(err), preview(r.Summary, 80), err, diag)
 		e.touch(t.ID)
 		return
 	}