feat(mybatis): add repository extension points for aggregates with child tables

[poppycoderr]

Background

GenericRepositoryImpl maps one aggregate to one row: E → P, one mapper call, P → E. An aggregate with child entities, such as an order with its lines, did not fit. A subclass had to override create, update, find, findAll, page and remove, and re-implement the optimistic-lock check and the event publishing that the base class does around them. This came up when starting the order context of the ddk-mall reference application.

Changes

• Four protected no-op methods on GenericRepositoryImpl, called around the root table's reads and writes in the same transaction:

create / createAll   insert root            → afterInsert(po)        → map to entity → publish events
update / updateAll   update root (version)  → afterUpdate(po)        → map to entity → publish events
                     0 rows                 → ConcurrentUpdateException, afterUpdate is not called
find / findAll / page   select roots        → afterLoad(pos)         → map to entities
remove / removeAll   beforeRemove(keys)     → delete roots

• The PO carries the children in a field marked @TableField(exist = false). The E → P converter fills it, afterInsert / afterUpdate write it to the child table, and afterLoad reads it back for the P → E converter. The aggregate is still saved and loaded as a whole; the application layer does not see the child table.
• afterLoad receives all roots of one query, so children can be fetched with a single IN query instead of one query per root.

Existing repositories are unaffected: the methods do nothing unless overridden.

---

背景

GenericRepositoryImpl 把一个聚合映射成一行：E → P，调用一次 Mapper，P → E。带子实体的聚合（例如订单和订单行）放不进这个模型。子类只能把 create、update、find、findAll、page、remove 全部覆盖，再把基类围绕它们做的乐观锁校验和事件发布重写一遍。这个问题是在开始做 ddk-mall 参考应用的订单上下文时遇到的。

改动

• GenericRepositoryImpl 新增四个默认为空的 protected 方法，在根表读写的前后调用，与根表在同一个事务里，调用时机见上文。
• PO 用标了 @TableField(exist = false) 的字段带上子对象：E → P 的转换器填好它，afterInsert / afterUpdate 把它写进子表，afterLoad 把它读回来交给 P → E 的转换器。聚合仍然作为一个整体保存和加载，应用层看不到子表。
• afterLoad 一次拿到本次查询的全部根对象，子表可以用一条 IN 查询取回，而不是每个根对象查一次。

现有的仓储不受影响：这些方法不覆盖就什么都不做。