validate could check hunks of an index in parallel

[sourcefrog]

Following on from #144, validate could check multiple hunks from an index in parallel. This would help performance when there are fewer than NCPUS versions, but each very large - for example after the first backup of a large tree.

The hunks can basically all be checked independently, except that we do also want to check the last apath in one hunk sorts before the first apath in the successor.