Remove mailbox for now

[sjmiller609]

Reverts #260

---

Note

Medium Risk
Changes networked fork/restore behavior on Firecracker by dropping the fast mailbox/ack path; restores depend on vsock guest-agent reconfigure, which affects timing and failure modes for forked standby restores.

Overview
Reverts the guest-initiated resume network mailbox flow (PR #260): the host no longer patches Firecracker snapshot memory, waits for guest UDP acks, or arms mailbox env on create/start.

Restore and fork networking after standby/running resume now rely only on the existing host-initiated path—reconfigureGuestNetwork via guest-agent gRPC (with shell ip exec fallback)—when a fresh allocation is applied for forked standby restores.

Removed code spans lib/mailbox, instance handoff helpers (prepareResumeNetworkHandoff, mailbox patching/UDP waiter), guest-agent VMGenID/mailbox watcher, related tests/docs, and a ForkSnapshot API unit test. Linux test env no longer passes HYPEMAN_TEST_NETWORK_TMPDIR; parallel test network locks/leases use the system temp dir only.

^{Reviewed by Cursor Bugbot for commit 71309cb. Bugbot is set up for automated code reviews on this repo. Configure here.}

[firetiger-agent]

Created a monitoring plan for this PR.

What this PR does: Removes the guest-initiated resume-network mailbox — a fast-path where hypeman pre-patched VM snapshot memory with new network identity before restore. Network reconfiguration after standby→running transitions now uses only the shell-exec path (already deployed as the fallback).

Intended effect:

• Spawn success rate: baseline 99.9–100% (310K–754K successes/hr active hours); confirmed if rate holds at ≥99.9%
• POST /browsers 500 errors: baseline 0–6/hr during clean active hours (10K–14K req/hr); confirmed if no sustained increase above ~50/hr post-deploy

Risks:

• Restore failures (exec path less reliable than mailbox) — kernel_invocation_spawn_total{success=false}, alert if >1,000/hr sustained for 2+ hours (clean baseline: 0–240/hr; infra-noise ceiling ~774/hr)
• Restore latency increase — POST /browsers p95 latency, alert if sustained above ~12s (baseline 8–10s p95)
• Fork readiness regression — spawn failures or "wait for forked guest agent readiness" ERROR logs; alert on any new occurrence
• failed to configure guest network after restore ERROR logs — any occurrence post-deploy is actionable

Status updates will be posted automatically on this PR as monitoring progresses.

View monitor

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 58ce89f. Configure here.}

[sjmiller609]

CI server firewall was messed up dropping guest -> host communication, which is why CI was failing:

TASK [nftables_firewall : Deploy nftables ruleset] ********************************************************************************************************************************
--- before: /etc/nftables.conf
+++ after: /Users/steven/.ansible/tmp/ansible-local-67508yu0zar6_/tmpbq_ycng0/nftables.conf.j2
@@ -20,6 +20,9 @@
...
+
+        # Trusted host-local interfaces
+        iifname "hm*" accept