How to prune the inverted index

[musalbas]

See #2615

[musalbas]

I think it's definitely possible to implement inverted index pruning purely in inverted_index.go by modifying AddMultihashesForShard and GetShardsForMultihash. Here's how:

• Let's say storage window is 30 days. We keep a badger index for the last 30 days of data; and one for the 30 days prior (60 days total).
• By construction, AddMultihashesForShard only ever needs to add indexes to blocks in the most recent badger index, thus no height parameter is needed in the call stack.
• GetShardsForMultihash needs to try both indexes anyway -- doing up to two reads.
• There should be some background routine that deletes the old index automatically once we've accumulated >30 days' worth of blocks.

[distractedm1nd]

@Wondertan is there any possibility of a CID being found in two distinct blocks? I don't think so but I may be forgetting something, like padding

[Wondertan]

In the case of padding, there is the same share in almost every block which will have the same indexed CID.