MirrorUP-v0.6.5.mcdr

修复

• 部分 region 的 chunk 永久漏同步：某个 region 在某一轮写失败（源端正在写、目标被占用）或被 !!mu cancel 取消后，下一轮的脏扫描下限已经越过那些 chunk 的时间戳，它们会被当成"没变"而不再写入。
• 现在每个 region 单独记录"上一次成功写入那一轮的开始时间"（索引 regions.floor，schema v6），失败不会推进它，下一轮会重新读取并补写。
• 整文件兜底（外置 chunk / 头损坏）时会清掉该 region 的 chunk 状态，下一轮整份重扫。

升级后请执行一次 !!mu sync --full，把之前可能已经漏掉的 chunk 补回来。

Fix

• Chunks of some regions were never synced again: when a region failed to write in one round (source being written, target busy) or the round was cancelled, the next round's dirty-scan floor had already moved past those chunks' timestamps, so they were treated as unchanged and skipped forever.
• Each region now records the start time of the last round in which it was successfully written (index regions.floor, schema v6). A failure does not advance it, so the next round re-reads and re-writes those chunks.
• When a region falls back to a whole-file copy (external chunks / corrupt header), its chunk state is dropped so the next round rescans it completely.

Run !!mu sync --full once after upgrading to repair chunks that may already have been missed.