fix: make CH pause/resume idempotent via error matching (no extra vm.info)

[CMGS]

Summary

• CH's Vm::pause / Vm::resume reject same-state transitions (vmm/src/vm.rs:412-440) — a cocoon process SIGKILLed mid-snapshot leaves the VM stuck Paused, after which every subsequent snapshot save fails with 500 InvalidStateTransition(Paused, Paused). This was the root cause of the journal noise that helped trigger fix: kill orphan VMM via /proc cmdline fallback when pidfile pre-removed #51's orphan chain on GKE prod.
• Cocoon-side fix: pauseVM / resumeVM swallow CH's exact Invalid transition: InvalidStateTransition(<state>, <state>) 500 body via APIError.Code + strings.Contains on the message. No extra vm.info round-trip.
• FC research confirms FC's pause/resume API is inherently idempotent at the vCPU event-loop level (src/vmm/src/vstate/vcpu.rs ack Pause from paused() and Resume from running() without error). The "Non-idempotent" comment in firecracker/api.go was wrong — comment-only cleanup.

Repro (testbed)

cocoon vm run --nics 0 --memory 512M ghcr.io/cocoonstack/cocoon/ubuntu:24.04
curl --unix-socket /var/lib/cocoon/run/cloudhypervisor/$VMID/api.sock -X PUT http://localhost/api/v1/vm.pause  # simulates stuck-paused
cocoon snapshot save --name test $VMID
# Before: pause: PUT vm.pause → 500: ["...InvalidStateTransition(Paused, Paused)..."], snapshot save fails
# After:  snapshot saved: <id>, VM resumes to Running

Test plan

• make lint 0 issues, both GOOS=linux and GOOS=darwin
• new TestIsAlreadyInStateError (9 cases: paused/running match, wrong-state, non-500, different-transition, non-APIError, nil, wrapped via fmt.Errorf %w, empty state)
• testbed e2e: manual vm.pause via curl → cocoon snapshot save succeeds (was failing pre-fix); cleanup leaves no orphan
• deploy GKE prod, watch journal for InvalidStateTransition(Paused, Paused) recurrence after next e2e batch

[Copilot]

Pull request overview

Makes Cloud Hypervisor pauseVM / resumeVM idempotent by detecting and swallowing CH's InvalidStateTransition(<state>, <state>) 500 response, so a stuck-Paused VM (e.g., after a SIGKILL mid-snapshot) can recover without an extra vm.info round trip. Also corrects an inaccurate "Non-idempotent" comment on the Firecracker pause/resume helpers, since FC's vCPU event loop already acks redundant transitions.

Changes:

• Add isAlreadyInStateError helper matching CH's exact 500 body via utils.APIError.Code + strings.Contains, and use it in pauseVM / resumeVM.
• New TestIsAlreadyInStateError covering 9 cases (matches, wrong state, non-500, non-APIError, nil, wrapped error, empty state, etc.).
• Update Firecracker pauseVM / resumeVM doc comments to reflect that FC's pause/resume is inherently idempotent.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
hypervisor/cloudhypervisor/helper.go	Adds error-matching helper and wires it into pauseVM/resumeVM to swallow same-state transitions.
hypervisor/cloudhypervisor/helper_test.go	New unit tests for isAlreadyInStateError.
hypervisor/firecracker/api.go	Corrects misleading "Non-idempotent" doc comments on FC pause/resume.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.