lock_move_reorganize_page() takes exclusive lock_sys.latch via LockMutexGuard

lock_move_reorganize_page() only touches record locks belonging to a single page, all of which live in one lock_sys.rec_hash cell. It nonetheless write-latches the global lock_sys.latch via LockMutexGuard, blocking every concurrent lock_sys.rd_lock() acquirer (e.g. lock_rec_lock(), lock_table()) server-wide for the whole reorganize (in case locks are present).

commit 983df00 downgrades LockMutexGuard to LockGuard (shared lock_sys.latch + exclusive latch on the page's cell). An exclusive cell latch is sufficient: it protects the cell's lock chain, the lock bitmaps, and lock->type_mode (including LOCK_WAIT). The per-iteration trx->mutex that already guards lock_rec_add_to_queue() continues to protect each trx's trx_locks list and lock_heap.