ci: fix arm64 QEMU block hotplug coverage

[fwerkor]

Summary

• add explicit PCIe root ports for arm64 QEMU block-device hotplug
• wait for exact guest output markers before host-side device actions
• fail on QEMU monitor errors and stale static block-device nodes
• only emit success markers after real ArgosFS operations and content checks succeed
• pin CapOS to the arm64 PCIe hotplug kernel fix from armsr: enable PCIe hotplug on armv8 CapOS#108

Root cause

The arm64 QEMU virt machine cannot hotplug a virtio-blk-pci device directly on pcie.0. The previous scripts ignored the monitor error, accepted static /dev/vd* nodes as proof of device presence, and could react to echoed marker commands before the guest actually executed them. CapOS also disabled CONFIG_HOTPLUG_PCI_PCIE, so devices attached through a root port were not enumerated by the guest.

Validation

• bash -n on all changed QEMU scripts
• shellcheck on all changed QEMU scripts
• git diff --check
• cargo test --workspace --all-targets — 156 passed
• direct arm64 QEMU verification of root-port block hotplug
• monitor helper rejects the former direct-root hotplug error
• exact marker test rejects terminal command echo and accepts real guest output
• pinned CapOS commit verified fetchable from a fresh repository

The full rebuilt CapOS arm64 image path is exercised by this PR CI.