Serialize/Deserialize an unfiltered row iterator.
 The serialization is composed of a header, follows by the rows and range tombstones of the iterator serialized
 until we read the end of the partition (see UnfilteredSerializer for details). The header itself
 is:
 
<cfid><key><flags><s_header>[<partition_deletion>][<static_row>][<row_estimate>]
 where:
     <cfid> is the table cfid.
     <key> is the partition key.
     <flags> contains bit flags. Each flag is set if it's corresponding bit is set. From rightmost
         bit to leftmost one, the flags are:
         - is empty: whether the iterator is empty. If so, nothing follows the <flags>
         - is reversed: whether the iterator is in reversed clustering order
         - has partition deletion: whether or not there is a <partition_deletion> following
         - has static row: whether or not there is a <static_row> following
         - has row estimate: whether or not there is a <row_estimate> following
     <s_header> is the {@code SerializationHeader}. It contains in particular the columns contains in the serialized
         iterator as well as other information necessary to decoding the serialized rows
         (see {@code SerializationHeader.Serializer for details}).
     <partition_deletion> is the deletion time for the partition (delta-encoded)
     <static_row> is the static row for this partition as serialized by UnfilteredSerializer.
     <row_estimate> is the (potentially estimated) number of rows serialized. This is only used for
         the purpose of sizing on the receiving end and should not be relied upon too strongly.
 
 Please note that the format described above is the on-wire format. On-disk, the format is basically the
 same, but the header is written once per sstable, not once per-partition. Further, the actual row and
 range tombstones are not written using this class, but rather by 
ColumnIndex.