P2P hardening: fix bucket refresh, pool contention, server RW timeouts, ignore filter & NodeList slice bug

Summary

In testnet, StoreArtefacts occasionally hung and supernodes gradually “lost” the network (failing to discover/reconnect). This PR addresses several correctness and resiliency issues in the P2P/Kademlia layer that could cause handler leaks, global stalls during handshakes, incorrect LRU maintenance, futile re-dials to ignored/self nodes, and malformed peer lists.

Changes

• Server I/O timeouts: add per-message SetReadDeadline/SetWriteDeadline (~30s) in handleConn
  File: supernode/p2p/kademlia/network.go
• Pool lock narrowing: stop holding connPoolMtx across NewSecureClientConn (TLS/TCP handshake) using double-check pattern
  File: supernode/p2p/kademlia/network.go
• Correct bucket refresh: call refreshNode with node.ID (raw), and no-op if not found to preserve LRU
  Files: supernode/p2p/kademlia/dht.go, supernode/p2p/kademlia/hashtable.go
• Ignore filter correctness: key ignoredMap by string(ID) and look up with the same representation; also filter includeNode
  File: supernode/p2p/kademlia/hashtable.go
• NodeList integrity: fix NodeIDs()/NodeIPs() to allocate with capacity and append (no double-length / empty entries)
  File: supernode/p2p/kademlia/node.go
• RPC timeouts (clarified): ensure explicit exec timeouts for slow paths (store/replicate/batch find)
  File: supernode/p2p/kademlia/network.go
• Minor defensive guards: nil-safety around response handling; comments and small hygiene where relevant

Why this helps StoreArtefacts

• Server RW deadlines prevent half-open/slow peers from pinning handler goroutines indefinitely (no FD/goroutine leak over time).
• Handshake outside pool lock prevents a single slow dial from globally blocking unrelated RPCs (parallel StoreArtefacts/finds keep flowing).
• Correct LRU refresh + working ignore filter improve target selection and avoid re-contacting self/banned nodes.
• Clean NodeList output removes subtle routing/replication skew from empty IDs.

Risk / Mitigations

• Timeouts are conservative and apply per message; they can be tuned if we see premature closes in high-latency environments.
• Pool double-check closes the losing connection to avoid leaks.
• No changes to protobuf or public SDK surfaces.

Test Plan

• Unit/integration (race):