Add multi-peer parallel block download during IBD #27

Summary

IBD currently uses a single peer, bottlenecking sync on network latency. This adds parallel block download from multiple peers while validating sequentially through process_block.

Proof-of-concept: https://github.com/pzafonte/kernel-node/tree/multi-peer-ibd

Design

• PeerManager spawns N threads (configurable --maxpeers, default 4) with auto-reconnect
• Shared download queue (Mutex<VecDeque<BlockHash>>) populated after header sync
• In-flight set (Mutex<HashSet<BlockHash>>) prevents duplicate requests across peers
• Per-peer state machine: AwaitingHeaders -> AwaitingInv / AwaitingBlock with 16-block batches
• Single validation thread via sync_channel(32) chainman is not thread-safe for concurrent process_block
• First peer syncs headers; late peers skip straight to block download (AtomicBool flag)
• Stalled peers (30s timeout) are disconnected; unreceived blocks re-enqueued for other peers
• Logarithmic block locators for getheaders/getblocks

This is a proof-of-concept implementation. Design tradeoffs favor simplicity over production-readiness. Each is documented with an inline // Tradeoff: comment explaining the cost and what a production implementation would do differently. There are probably other enhancements that could be made too.

Test Coverage

27 tests: 21 unit tests (batch popping, message construction, deduplication, state machine) + 6 integration tests with regtest ChainstateManager.