OpenVMM VMM test: enable openvmm_linux_x64_boot

[ppenna]

Goal

Enable the OpenVMM VMM test openvmm_linux_x64_boot in NVX's existing KVM, MSHV, and WHP OpenVMM test matrix.

Coverage added

Adds the canonical Linux-direct boot, pipette connection, clean guest poweroff, and VMM teardown smoke test used by upstream.

Source: openvmm\vmm_tests\vmm_tests\tests\tests\multiarch.rs.

Upstream evidence

The corresponding canonical test passed on Linux/KVM, Linux/MSHV, and Windows/WHP in microsoft/openvmm workflow run 35255255183 on September 17, 2026. NVX currently executes only openvmm_microvm_test_pvh_x64_phase_1_lifecycle and test_ttrpc_microvm_pvh_snapshot; neither canonical test exists in Microsoft upstream at commit e7de393, so this issue adds complementary upstream coverage rather than duplicating the current fork-only checks.

The nanvix/openvmm version may contain additional fork behavior. Treat the first backend-complete run as canary evidence and diagnose any difference rather than weakening or silently skipping the test.

Required artifacts and prerequisites

Requires OpenVMM, the packaged Linux-direct test kernel/initrd, and a locally built pipette guest agent. These are handled by cargo xflowey vmm-tests-run; no full guest-OS VHD is required.

The issue should not introduce a dependency on the multi-gigabyte Windows, Ubuntu, Alpine, or FreeBSD guest-image set used by the full upstream standard suite.

Implementation notes

• Add test(openvmm_linux_x64_boot) to the documented selector set used by OPENVMM_MICROVM_TEST_FILTER in scripts\nvx_tools\ci.py, preserving the two existing tests.
• Refactor the filter into named artifact-class groups if needed so future additions remain reviewable.
• Update scripts\test_nvx_tools.py to prove the selector is present and the existing --release --ci-profile --skip-vhd-prompt behavior remains intact.
• Ensure Flowey installs/builds the required Rust targets and artifacts on Linux and Windows without relying on preexisting runner state.
• Reuse the shared OpenVMM input/cache work from Improve CI reliability with shared OpenVMM input caches and binary artifacts #73 and Improve CI reliability with shared OpenVMM caches and artifacts #76; cache misses must remain correct.
• Keep the existing backend validation and failure-log upload behavior.

Acceptance criteria

• The exact test openvmm_linux_x64_boot is executed, not skipped, on Linux/KVM.
• The exact test openvmm_linux_x64_boot is executed, not skipped, on Linux/MSHV.
• The exact test openvmm_linux_x64_boot is executed, not skipped, on Windows/WHP.
• Existing fork-specific PVH lifecycle and ttrpc snapshot tests still pass on all three backends.
• Cold-cache and warm-cache CI runs both succeed.
• Failure logs identify the test and preserve Petri/OpenVMM diagnostics.
• python scripts\nvx.py verify and the directly affected NVX Python tests pass.

Rollout

• If the first matrix run exposes a fork-specific instability, land it as an explicitly visible non-blocking canary only with a linked follow-up; promote it to required coverage before closing this issue.