fix potential panic which may happen when subscribe the region and meet rollback and prewrite entry

The bug may occurs when:
1. T1 prewrites and commits key K (start_ts=100, commit_ts=110)
    → CF_WRITE[K@110] contains T1's committed write

2. T2 prewrites key K as a SECONDARY key (start_ts=111), and CDC subscribe the region 
   → CF_LOCK[K] contains T2's lock
   → CF_LOCK[K@111] scanned out when doing incremental scan by the `scan_locks_from_storage`
   → CDC incremental scan finished and set to the resolver by the `finish_scan_locks`

4. T2 rolls back key K
   → Finds overlapped_write from T1 at commit_ts=110
   → Since K is NOT T2's primary: protected=false
   → make_rollback() returns None (no CF_WRITE entry!)
   → Only deletes from CF_LOCK
   → CDC ignores the DELETE operation

5. T3 prewrites key K
   → Triggers resolve_lock which removes T2's stale lock from CF_LOCK
   → No CDC notification

6. CDC processes T3's prewrite
   → Calls push_lock for T3
   → Finds T2's stale lock still in lock_tracker
   → Assertion fails: different start_ts

fix potential panic which may happen when subscribe the region and meet rollback and prewrite entry