fix(split): add flush_memtable for copy_checkpoint_to_dir

[hycdong]

In pegasus pr 535, we use rocksdb's original API CreateCheckpoint to do checkpoint. In function copy_checkpoint_to_dir_unsafe, it will create a checkpoint without flush memtable, code like below:

// CreateCheckpoint() will not flush memtable when log_size_for_flush = max
status = chkpt->CreateCheckpoint(checkpoint_dir, std::numeric_limits<uint64_t>::max());

When parent partition prepare states for child to learned, it should create a checkpoint with memtable flushed. As a result, this checkpoint will contain more data, less data will be learned during async-learn stage.

This pull request adds flush_memtable option for function copy_checkpoint_to_dir, the pegasus implementation is in pegasus pr 684.