Defragmentation policy should issue dataSize commands with estimate true

Currently, the defragmentation policy is issuing datasize commands with estimate set to false. This can make the operation very slow for large chunks. We should change this field to true.

We also need to ensure that potential errors caused by incorrect datasize after this change are handled - for example, if the defragmentation tries to issue a move for a chunk that is too large, we should ensure the defragmenter is not continuously retrying this operation.