validate should check indexes in parallel

[sourcefrog]

Validate currently checks indexes purely sequentially, but it would be faster to parallelize:

1. check multiple indexes in parallel
2. check multiple hunks from inside each index in parallel

In principle we could also parallelize checking different entries from inside a hunk, but this is probably not useful. Most of the time is spent getting the hunk from disk, decompressing, and deserializing it, and the per-entry work is very small.

Probably this is better to do after #141, the fix for which involves changing index validation to return a set of referenced addresses.

---

Checking multiple hunks inside an index in parallel is possible, but a little more complicated because we also want to check ordering constraints between hunks, so I deferred that to #145.