Fix memory leak when calling get_writebatch and avoid unnecessary clones

[jkurian]

Memory leak

It seems wi is not deallocated properly which causes a memory leak when calling get_writebatch.

rocksdb_transaction_get_writebatch_wi allocates memory with malloc, which never gets deallocated.

There is rocksdb_writebatch_wi_destroy but that tries to delete the memory allocated with malloc which is UB and tries to clear rep. Using rocksdb_free seems to be the right choice.

Extra allocations

There also seems to be unnecessary cloning of the serialized writebatch. The code clones the data once by calling to_owned on the slice but the data is also cloned here:
https://github.com/facebook/rocksdb/blob/main/db/c.cc#L1905

To improve that, I think we can just pass the ptr directly to rocksdb_writebatch_create_from

[stanislav-tkach]

Thanks for the pull request! Would you mind creating an issue about rocksdb_writebatch_wi_destroy in the rocksdb repository?