Parallel key scraping

[liamaharon]

Original issue paritytech/polkadot-sdk#174

---

Motivation

The current key scraping implementation is is quite inefficient, fetching 1k keys at a time, waiting for each batch of 1k keys to be returned before asking for the next one

https://github.com/paritytech/substrate/blob/master/utils/frame/remote-externalities/src/lib.rs#L348-L392

This is not so bad when fetching keys from a local node where latency is not a big factor, but can slow things down very significantly when using a remote node when most of the time spend fetching keys becomes latency rather than the node doing work.

Unfortunately, most nodes seem to have a hard limit imposed preventing more than 1k keys being fetched in a single request.

Suggested solution

Rather than fetching all keys for the prefix in a single rpc_get_keys_paged request here https://github.com/paritytech/substrate/blob/master/utils/frame/remote-externalities/src/lib.rs#L528-L533, split the prefix up into N parts and run the rpc_get_keys_paged tasks in parallel using async/await.

e.g. two parallel tasks:
start_keys: 0x000000 end_keys: 0x777777
start_keys: 0x777777 end_keys: 0xffffff

Bonus (probably for a seperate issue/PR once this one is closed)

Fetch storage values for keys as they're pulled from the node, instead of waiting for all keys to be pulled before fetching values.

[liamaharon]

actually going to leave this in substrate as a remote-externalities issue.