virtualbox: VBoxVMI batched physical writes

[Wenzel]

Closes #285.

• VBoxVMI driver implements write_physical through batched PHYS_WRITE requests (multiple ranges, per-page status) via rs-vboxvmi 9a349ca. Entire range set is prevalidated; completed-batch statuses survive a later malformed reply. Writes are non-atomic across pages; bytes_read/errors identify the readable/writable prefix consistently.
• Existing core-ground-truth write test now runs under both methods. A VM-state guard is armed before pause and all fallible page-selection/core operations; the byte-restoration guard drops before resume.
• Driver/test docs: both methods now support physical writes.

Server: Wenzel/virtualbox#8 (merged into vmi/vboxvmi).

Verified live: VBoxVMI 512 MiB (seed 1791797247855151235; pages 0xf000, 0xffff000, 0x1ffff000) and 5120 MiB (seed 1791798462874390042; pages 0xf000, 0xaffff000, 0xdffff000, 0x15fff9000 >4 GiB), each checked in fresh cores before/after resume and after restore. Full VBoxVMI regression and MachineDebugger write/core pass. fmt/clippy/tests green locally.