Implement PeerDAS

[nalepae]

Remaining tasks:

• Initial sync: Backfill data columns when starting from a checkpoint.
• Initial sync: Do not request more than 64 data columns: Request 64 data columns then reconstruct the rest.
• Implement validator custody.
• Peer sampling with trailing fork choice: Integrate the output of sampling in the forkchoice?
• Check why we cannot run initial sync when starting a node during epoch 0, in slot [1, 31]. (Starting a node since genesis works, starting a node from epoch 1 and later works.) Note: This issue may be already here with blobs (without data columns.)
• Indroduce a nasty peer flag that holds AND/OR (both are important) alterate data columns on RPC responses.
• Add unit tests where they are not yet added.
• Replace the cKZG library by the goKZGone (when available) - In progress.
• ⚠️ Implement number of blobs as a CL parameter. ⚠️
• Decouple network subnets from Core methods as referenced.
• Peer sampling: Find more peers if a column has no corresponding peers. (Can do like in internalBroadcastDataColumn)
• Add PeerDAS specific metrics
• Clarify usage of --minimum-peers-per-subnet=<n>
• Should we downscore peer on sampling? (Maybe the peer does not see the block)
• Should we stop using a peer when failed on sampling? (To avoid DDoS)
• Check metadataV3 interop with other clients if before EIP-7594
• Refactor code where GetValidCustodyPeers is used. ==> We need to fetch data from peers that are not super nodes.
• prysmctl: Implement byRoot and byRange data columns requests
• Using subnet sampling only: Should, when using data columns by root requests (and also by range, during initial sync), the node request only custody columns, or custody columns + extra subnet sampling columns? If requesting only custody columns, then a strange situation arises: When receiving a block with a extra subnet sampling column missing, then the node will ignore the block. A few sec after, when requesting the blocks and the associated columns, the node won't need the subnet sampling extra columns to import the block.
• Find why sometimes ENR records are not correctly updated via discovery.
• Find why, with Grandine in the mix, discv5 iterator.Next is soooo slow.
• Test with only one bootnode (and not all nodes)
• By range (and by root) requests: Run verifications for finalized blocks?
• ByRoot requests: choose all peers like on ByRange requests, not only super nodes.
• ENR: Are we sure the sequence number cannot go backward after a reboot? (This trigs downscoring on Prysm.)
• subscribeWithParameters: Should we subscribe to subnets / find peers 1 epoch in advance for getSubnetsToSubscribe and getSubnetsToFindPeersOnly?
• Subnets unsubscribing: Handle the same way dynamic (fork epoch) and non dynamic (fork epoch + 1) unsubscriptions.
• Re-implement part of code where DataColumnsAdmissibleCustodyPeers and DataColumnsAdmissibleSubnetSamplingPeers are needed. We don't want to use only "superset peers".
• What happens if a peer announces less custody groups via Metadata/ENR than the minimal requirement? Ban?