Replace GetNode heartbeat hack with proper Heartbeat RPC

[shiv-tyagi]

The agent heartbeat was piggybacking on GetNode RPCs, which gave GetNode a write side-effect and discarded the telemetry the agent collected (always sending 0,0 for cpu_load/free_memory_mb).

• Add unary Heartbeat RPC (replaces both the GetNode hack and the unused streaming Heartbeat stub) that carries real telemetry
• Remove write side-effect from GetNode (now a pure read)
• Return NotFound for heartbeats from unregistered nodes so agents get visibility into controller state mismatches
• Remove inline string-matching recovery from update_heartbeat; node recovery is handled solely by check_node_health via the state machine (admin_locked flag, not reason strings)
• Delete dead WalOperation::NodeHeartbeat (never proposed, heartbeat telemetry is ephemeral and does not belong in the WAL)
• Remove redundant tests covered by existing state machine and health check unit tests
• Remove unused tokio-stream dependency and streaming stub from spurctld
• Update spur-k8s heartbeat to use the new RPC; remove the re-register-on-NotFound fallback (Raft WAL preserves node state across restarts, registration is the node watcher's job)