Split the chunk offset tables to reduce memory usage while dumping

[ph0llux]

To reduce the memory usage while dumping a very large amount of data (e.g. a physical disk with 20TB) to a single segment, the chunk offset table should be split in multiple tables - this tables should be written periodically into the Zff container (currently, the full table is cached in memory while dumping the data - a single table entry needs 2*8 bytes, so you need 500MB memory space for each TB of data by using a chunk size of 32kB).
At the end of the segment there should be an additional table which contains the appropriate offsets to the chunk-offset tables.
Due this tables are sorted HashMaps (or BTreeMaps), they can be variable in their size.