[BUG] - A relay only downloads one block at the time in Blockfetch deadline mode.

[karknu]

Internal

Describe the feature you'd like
The following scenario has been observed on the Shelley testnet.

Node A, IOHK and C are close, node B is far away.

1. A header for a new block, H_n from node A arrives at IOHK’s relay
2. It is for a new slot so IOHK’s relay decides to download it
3. A header for a new block H_n’ from node B arrives at IOHK’s relay, it is
   for the same slot but with a lower vrf value.
   IOHK’s relay knows that it will switch to it but it can’t start the download
   because it is limited by bfcMaxConcurrencyDeadline.
4. Block H_n arrives, and is adopted by IOHK’s relay so it announces it
   to all its peers, even if it knows it would like to switch ASAP.
5. Node C learns of H_n from IOHK and ask for it.
6. Block H_n arrives at Node C and is adopted.
7. Node C produces a new block
8. Block H_n’ arrives at IOHK’s relay. It switches and announces it to
   its peers.

Make bfcMaxConcurrencyDeadline and bfcMaxConcurrencyBulkSync configurable.
bfcMaxConcurrencyDeadline of one will work fine for normal clients (Daedalus) but should be set to two or more for pool relays.

Describe alternatives you've considered
One could increase the hardcoded bfcMaxConcurrencyDeadline in ouroboros-network, but it that would mean that it is active for all nodes.

[karknu]

Depends on IntersectMBO/ouroboros-network#2363