fix(memory): 定宽时间戳存储 + rowid 确定性排序 + 遗留数据归一化

- formatTime 改用固定 30 字节布局(9 位小数秒)，让 TEXT 列的字典序与时间序一致（RFC3339Nano 会去掉尾随零产生变宽字符串，秒内会反序）
- memory_items / project / project_chapter 的 ORDER BY 全部补 rowid 兜底，消除同 tick 打平时的顺序不确定
- 开库时幂等归一化变宽的历史值；无法解析的行原样保留，不阻塞开库
- 回归测试：变宽排序反例、同一时间戳插入序、归一化幂等与跳过

internal/runtime/memory/store.go

@@ -158,7 +158,8 @@ type rowScanner interface {
 // db must be a live, migrated connection (the migration framework itself is
 // owned by internal/db). A nil db returns ErrNilDB. log may be nil, in which
 // case a discard logger is used. ctx controls schema creation and must be
-// honored by the caller.
+// honored by the caller. Any legacy variable-width timestamps are normalized in
+// place (see normalizeMemoryTimestamps), so text ordering matches time ordering.
 func NewStore(ctx context.Context, db *sql.DB, log *slog.Logger) (*Store, error) {
 	if db == nil {
 		return nil, ErrNilDB
@@ -176,6 +177,9 @@ func NewStore(ctx context.Context, db *sql.DB, log *slog.Logger) (*Store, error)
 	if err := ensureProjectIDColumn(ctx, db); err != nil {
 		return nil, err
 	}
+	if err := normalizeMemoryTimestamps(ctx, db); err != nil {
+		return nil, err
+	}
 	return &Store{db: db, log: log}, nil
 }
 
@@ -227,11 +231,12 @@ func (s *Store) Get(ctx context.Context, id string) (MemoryItem, error) {
 	return item, nil
 }
 
-// ListByPersona returns every item for personaID, newest first. It is the
-// persona-isolation boundary: items belonging to other personas are never
-// returned, so personas never share memory (design §4). Use ListScope when a
-// caller needs to see shared user/project scopes or another persona's read-only
-// items.
+// ListByPersona returns every item for personaID, newest first. rowid breaks
+// created_at ties so items written within the same clock tick keep a stable
+// newest-first order. It is the persona-isolation boundary: items belonging to
+// other personas are never returned, so personas never share memory (design
+// §4). Use ListScope when a caller needs to see shared user/project scopes or
+// another persona's read-only items.
 func (s *Store) ListByPersona(ctx context.Context, personaID string) ([]MemoryItem, error) {
 	if strings.TrimSpace(personaID) == "" {
 		return nil, ErrMissingPersona
@@ -239,22 +244,24 @@ func (s *Store) ListByPersona(ctx context.Context, personaID string) ([]MemoryIt
 	rows, err := s.db.QueryContext(ctx, `SELECT id, persona_id, role_id, project_id, scope, kind,
 		content, tags, confidence, source_conversation, created_at, updated_at,
 		decay_weight FROM memory_items WHERE persona_id = ?
-		ORDER BY created_at DESC`, personaID)
+		ORDER BY created_at DESC, rowid DESC`, personaID)
 	if err != nil {
 		return nil, fmt.Errorf("memory: list persona items: %w", err)
 	}
 	return scanItems(rows, "persona item")
 }
 
-// List returns every memory item, newest first, across all personas. It is the
-// aggregate/admin view that enumerates all persisted memory regardless of the
-// persona-isolation boundary; callers that must respect isolation should use
-// ListByPersona or ListScope instead. The returned slice is non-nil.
+// List returns every memory item, newest first, across all personas. rowid
+// breaks created_at ties so items written within the same clock tick keep a
+// stable newest-first order. It is the aggregate/admin view that enumerates all
+// persisted memory regardless of the persona-isolation boundary; callers that
+// must respect isolation should use ListByPersona or ListScope instead. The
+// returned slice is non-nil.
 func (s *Store) List(ctx context.Context) ([]MemoryItem, error) {
 	rows, err := s.db.QueryContext(ctx, `SELECT id, persona_id, role_id, project_id, scope, kind,
 		content, tags, confidence, source_conversation, created_at, updated_at,
 		decay_weight FROM memory_items
-		ORDER BY created_at DESC`)
+		ORDER BY created_at DESC, rowid DESC`)
 	if err != nil {
 		return nil, fmt.Errorf("memory: list items: %w", err)
 	}
@@ -378,7 +385,8 @@ func applyTimestamps(item MemoryItem) (MemoryItem, error) {
 }
 
 // scanItem reads one memory_items row. tags is stored as a JSON array string;
-// timestamps are stored as RFC3339 strings.
+// timestamps are stored as fixed-width UTC strings (timeLayout) so text
+// comparison orders them chronologically.
 func scanItem(sc rowScanner) (MemoryItem, error) {
 	var (
 		it                   MemoryItem
@@ -421,24 +429,6 @@ func newID() (string, error) {
 	return hex.EncodeToString(b), nil
 }
 
-// formatTime encodes t in UTC as an RFC3339 string. The value is always written
-// as UTC so ordering and round-trips are stable across processes.
-func formatTime(t time.Time) string {
-	return t.UTC().Format(time.RFC3339Nano)
-}
-
-// parseTime decodes a stored RFC3339 string. Blank input yields a zero time.
-func parseTime(s string) (time.Time, error) {
-	if strings.TrimSpace(s) == "" {
-		return time.Time{}, nil
-	}
-	t, err := time.Parse(time.RFC3339Nano, s)
-	if err != nil {
-		return time.Time{}, fmt.Errorf("memory: parse time %q: %w", s, err)
-	}
-	return t, nil
-}
-
 // discardLogger returns log if non-nil, otherwise a logger that writes nowhere.
 func discardLogger(log *slog.Logger) *slog.Logger {
 	if log != nil {