Bulk import scans all table metadata when removing load flags.

[keith-turner]

This is follow up work to #436. When bulk import is complete, load flags are removed. To do this all table metadata is scanned. Could just scan the metadata range that was bulk imported to.

[keith-turner]

I looked into doing this. I was thinking I could just use the min and max rows from the load mapping to scan the metadata table. However if a tablet at the min or max row has split then it may copy some load flags outside of the range. There are three possible ways to avoid this :

1. Ignore this edge case because metadata compaction will eventually clean up load flags.
2. Make the split code smarter about copying load flags. Currently all load flags go to both children, even if the loaded file does not. This may be impossible though for the case the file was compacted away.
3. Get the min and max row in the metadata table at the beginning of the fate operation and use these rows to limit the cleanup at the end. Since the min and max would be recorded before anything is loaded splits occurring after/during the fate load step would fall within this range. The could be done in the first fate step when it checks for merges.

Option 3 allows the final fate step to avoid opening the load mapping file to find the min and max row.