feat: .getMany: improve performance for many small blocks

[pgte]

Currently, in peer-crdt, the latency of replication of logs is bound to the latency of getMany(cids) when the blocks are not present.

Also, I think we should create benchmarks to have a baseline we can try to improve on.

[daviddias]

@pgte what's the next step now that #164 is done?

[pgte]

Currently, the greatest blocker is the overhead of useless connections.
There's nothing obvious that can be done at this level for individual bitswap exchanges.
After connection management gets solved, my informed guess for the next candidate is overhead on resolving through DHT.

[daviddias]

The .getMany call is also needed for ipfs/js-ipfs#856. The next step is really augmenting the tests -- https://github.com/ipfs/js-ipfs-bitswap/blob/master/test/bitswap.js#L102-L150 -- and testing perf.