Make P1 requests concurrent

[shawn-zil]

Treat it as just larger segments. Scheme:

1. When request_missing_metadata() is called, fire off batch request for different segments.
2. Insert peer into request queue back.
3. If V2 peer, keep firing batch until N peers used.
4. If V1 peer, STOP - we don't speed-up V1.

When responses are received. Scheme:

1. Insert response into a response cache.
2. Process in-order, based on request queue.
3. If response is Error/Empty, fire one request, insert into front of queue.
4. If response is valid, process the segment.
5. Repeat processing front-of-queue until done or waiting on response.
6. If queue is empty i.e. done, fire next batch request.
7. If history is hit, drop all pending requests.

When empty/error, mark that in the response cache.

Important to ensure that each peer is used only one-at-a-time.