Improve open-to-close semantics

This bugfix improves xrootd-hdfs's open-to-close semantics. We now synchronize before close, which appears to help force the namenode to update the metadata as soon as the file handle is closed.

Further, if the parsing checksums fails (for example, if the checksum writing is incomplete when the checksum request is done), then we now will retry with a new checksum calculation.