feat(network): Batches and parallelizes block downloads

[arya2]

Motivation

We want to improve Zebra's initial sync performance.

Solution

See commit messages for more information

Tests

Running full Testnet sync locally.

Follow-up Work

• Sync-specific peer quality tracking, P2C uses EWMA latency for load, but during sync what matters is throughput and reliability
• Switch from TCP to QUIC
• Peer chain height is known but unused for sync routing
• Peers only process one request at a time, but Zebra could send several requests at once rather than sending once, waiting for a response, etc.

AI Disclosure

• No AI tools were used in this PR
• AI tools were used: All changes were Opus 4.6 except updating the latency threshold from 0.95 to 0.9 somewhere.

PR Checklist

• The PR name is suitable for the release notes.
• The PR follows the contribution guidelines.
• This change was discussed in an issue or with the team beforehand.
• The solution is tested.
• The documentation and changelogs are up to date.

[Copilot]

Pull request overview

This PR improves Zebra's initial block sync performance by batching and parallelizing block downloads. Instead of requesting blocks one-at-a-time sequentially, the syncer now groups blocks into adaptively-sized batches, sends multiple batches concurrently, and overlaps download progress with tip discovery. This reduces per-request overhead, maximizes network utilization, and eliminates pipeline stalls. The PR also improves peer routing during sync: multi-block requests target peers at or above the current chain height, and tip discovery requests spread fanout across distinct peers for better diversity.

Changes:

• Implement batched block downloads with per-block retry logic and dynamic batch sizing based on observed block sizes
• Run multiple batches concurrently using SelectAll to interleave block arrivals and prevent waiting for entire batches
• Extend tips proactively while downloads are in-flight, and drain completed downloads during tip discovery, eliminating pipeline stalls
• Add peer routing logic to prefer peers at/above chain tip height for multi-block requests and distinct peers for tip-discovery fanout
• Update integration tests to verify batched request behavior and concurrent verification

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
zebrad/src/components/sync/downloads.rs	Implements batched download logic with dynamic batch sizing, retry mechanism, and block size tracking for adaptive batching
zebrad/src/components/sync.rs	Integrates batched downloads, adds drain_ready_downloads() to pipeline during tip discovery, and refactors tip extension to run concurrently
zebra-network/src/peer_set/set.rs	Adds peer routing for multi-block requests (targeting tall peers) and tip-discovery requests (targeting distinct peers), with recent-peer tracking
zebra-network/src/peer/load_tracked_client.rs	Adds remote_height() method to expose peer's reported chain height for peer selection during sync
zebrad/src/components/sync/tests/vectors.rs	Updates test expectations to verify batched request behavior instead of sequential per-block requests

[arya2]

Record the number of every hundredth block. Start at a random number in 0..99, i.e. compute and send the serialized size if block.height + random_num % 100 == 0