Speed up post-restore guest network reconfigure

[sjmiller609]

Summary

• add a guest-agent ReconfigureNetwork RPC that applies restored fork MAC/IP/route updates via netlink instead of spawning shell + ip commands
• use the RPC for post-resume reconfigure_guest_network, with fallback to the existing shell exec path for older guest agents/snapshots
• add a gated Firecracker perf harness for repeated snapshot-fork restore timing

Deft validation

On deft-kernel-dev from a short remote workspace (~/hmge):

• make build-embedded
• go test ./lib/guest ./lib/system/guest_agent
• go test ./lib/instances -run "TestGuestNetworkReconfigureConfig|TestGuestNetworkReconfigureCommand|TestNetworkConfigFromAllocation|TestRequiresRestoreConfigDiskRefresh" -count=1
• sudo env ... HYPEMAN_RUN_RECONFIGURE_GUEST_NETWORK_PERF=1 HYPEMAN_RECONFIGURE_GUEST_NETWORK_PERF_ITERS=3 /tmp/reconfigure-perf.test -test.v -test.run TestReconfigureGuestNetworkPerf -test.count=1 -test.timeout=12m
• sudo env ... go test ./lib/instances -run TestFirecrackerForkFromRunningNetwork -count=1 -timeout=12m -v

Perf notes

Baseline reproduced reconfigure_guest_network around ~797-943ms, dominated by first post-resume guest exec stream contact and shell/process execution. The netlink RPC reduced the step to ~197-214ms in the final 3-iteration run, with the successful unary apply taking ~35-53ms. Remaining cost is still first-contact grpc_unavailable after resume, so sub-100ms likely needs a separate guest-agent/vsock readiness change or guest-side proactive reconfigure.

[sjmiller609]

Closing in favor of #253, which combines the main ReconfigureNetwork + mailbox/VMGenID path with the exploratory pieces trimmed out. Keeping the branch for reference.