Use a call proof in the Grandpa warp syncing instead of multiple storage proofs

[tomaka]

At the moment, the warp syncing code calls the runtime, then sends one storage proof every time the runtime accesses the storage.

This means a lot of redundancy in the proofs, but more importantly a lot of network round trips.
Right now on a simple dev chain we do 12 round trips just for that. If the ping is 100ms, that means it takes 1.2 seconds instead of only 100ms.

[tomaka]

For context, the reason why it was done with simple storage proofs is that I expected one or two storage items to be accessed, rather than 12. At the time, call proofs weren't properly implemented.

[tomaka]

At the moment, warp syncing Westend (with a moderately old checkpoint) on my local machine takes about 2.5 to 5 seconds.
In my work-in-progress branch, this is now down to 1.5 to 2 seconds.