[network] Support multiple connections per peer

[zekun000]

Summary

• Add max_connections_per_peer config parameter to allow multiple TCP connections between the same two peers for increased throughput and reduced head-of-line blocking
• Add enable_active_multi_connection_dialing config parameter for safe rollout:
  • When false (default): node accepts incoming connections up to max_connections_per_peer but does not actively dial for additional connections
  • When true: connectivity manager actively dials until max_connections_per_peer is reached
• PeerManager now tracks connections by ConnectionId with round-robin selection for load balancing across connections
• Tie-breaking logic when at max connections favors inbound connections (disconnect outbound)
• NewPeer/LostPeer notifications only sent for first/last connection to maintain application compatibility
• Default values: max_connections_per_peer=1, enable_active_multi_connection_dialing=false for backward compatibility

Safe Rollout Strategy

1. First phase: Set max_connections_per_peer > 1 with enable_active_multi_connection_dialing=false
   • Nodes will passively accept multiple connections from peers that actively dial
   • Validates that PeerManager correctly handles multiple connections
2. Second phase: Enable enable_active_multi_connection_dialing=true
   • Connectivity manager will actively establish multiple connections per peer

Test plan

• Added E2E smoke test test_multi_connection_per_peer_validators - 2 validators with active dialing enabled
• Added E2E smoke test test_multi_connection_allows_simultaneous_dial - 4 validators with active dialing enabled
• Added E2E smoke test test_default_single_connection_per_peer - verifies default behavior unchanged
• Manual testing on devnet/testnet

Generated with Claude Code

[github-actions]

✅ Forge suite realistic_env_max_load success on c5483dab7f854496847f4573b633f8458cc3060c

two traffics test: inner traffic : committed: 13536.82 txn/s, submitted: 13536.93 txn/s, expired: 0.11 txn/s, latency: 2784.28 ms, (p50: 2700 ms, p70: 2900, p90: 3000 ms, p99: 3600 ms), latency samples: 5030460
two traffics test : committed: 100.02 txn/s, latency: 757.53 ms, (p50: 700 ms, p70: 800, p90: 900 ms, p99: 1500 ms), latency samples: 1720
Latency breakdown for phase 0: ["MempoolToBlockCreation: max: 2.292, avg: 2.182", "ConsensusProposalToOrdered: max: 0.169, avg: 0.166", "ConsensusOrderedToCommit: max: 0.046, avg: 0.043", "ConsensusProposalToCommit: max: 0.214, avg: 0.209"]
Max non-epoch-change gap was: 0 rounds at version 0 (avg 0.00) [limit 4], 0.51s no progress at version 36812 (avg 0.07s) [limit 15].
Max epoch-change gap was: 0 rounds at version 0 (avg 0.00) [limit 4], 0.30s no progress at version 2461454 (avg 0.30s) [limit 16].
Test Ok

• Grafana dashboard
• Humio Logs
• Axiom Logs
• Validator CPU Profile
• Fullnode CPU Profile
• Test runner output
• Test run is land-blocking