_ma_apply_redo_index: KEY_OP_DEL_PREFIX subtracts unchecked length from page_length (size underflow / huge bmove)

In `storage/maria/ma_key_recover.c`, `_ma_apply_redo_index()` handles `KEY_OP_DEL_PREFIX` as:
```c
case KEY_OP_DEL_PREFIX:
uint length= uint2korr(header);
header+= 2;
DBUG_ASSERT(length <= page_length - keypage_header);
bmove(buff + keypage_header, buff + keypage_header + length,
page_length - keypage_header - length);
page_length-= length;
```
In release builds (`DBUG_OFF`), the assert is a no-op. If redo supplies `length > page_length - keypage_header`, then:

• `page_length - keypage_header - length` underflows (unsigned)
• `bmove` is called with a huge size
• `page_length-= length` also wraps
  A corrupt `LOGREC_REDO_INDEX` applied during Aria recovery can therefore trigger a large out-of-bounds move.
  1. Impact
     Crash recovery / redo apply can crash or corrupt memory on a malformed DEL_PREFIX redo record.