vms-ddc: probe non-virtio (SCSI/NVMe) disks so SHOW DEVICE is faithful on bare metal

[baron-3dl]

What / Why

vms_devtab_probe_disks() (src/kernel-core/vms_devtab.c) enumerated only
/dev/vd* (virtio-blk). On bare metal / KVM -- OVMX's Rule-9 target -- the
system disk is SATA/SCSI or NVMe, so no disk unit was entered and
SHOW DEVICE VDA0: returned %SYSTEM-W-NOSUCHDEV (the operator's
direct-test finding, vms-ddc / vms-47d).

Three layers, so the fix is real end to end:

• executive: vms_devtab_probe_disks() now also probes /dev/sd* ->
  SDAn: and /dev/nvmeNn1 -> NVMEn:, entered via the existing
  vms_devtab_add_disk() path. INV-6 preserved exactly as the virtio loop:
  a unit is entered ONLY when exec_blockdev_lookup() actually resolves
  the node -- never a speculative row.
• kernel config: distro/kernel/ovmx-x86_64.config enables virtio-scsi
  built-in (CONFIG_SCSI/BLK_DEV_SD/SCSI_LOWLEVEL/SCSI_VIRTIO, =y
  -- the initramfs loads no modules). A prior attempt (reverted,
  9102593) used NVMe here, but the minimal from-source config lacks the
  MSI-X plumbing NVMe needs to bring up /dev/nvme0n1 in the guest.
  virtio-scsi reuses the already-enabled virtio transport and presents
  disks as /dev/sd*, exercising the SAME non-virtio probe path reliably
  in a minimal kernel.
• test harness: run_tests.sh attaches ONE real virtio-scsi disk
  (virtio-scsi-pci + scsi-hd, presenting /dev/sda).

Anti-LARP proof

test_kmod_disk.c asserts, against a REAL vms.ko in QEMU with a real
virtio-scsi disk attached:

• SDA0: exists in the executive's table (SS$_NORMAL), backing device
  sda
• SDA0:'s backing dev_t matches /dev/sda's independently
  stat()'d major:minor -- the same "no way to fake it" oracle the
  VDA0:..VDA400: checks use
• SDA100: (a second, absent SCSI unit) reports SS$_NOSUCHDEV (negctl)

Local verification

Kernel module (vms.ko, all TUs including the modified vms_devtab.c)
built cleanly against real kernel headers (make -C /lib/modules/$(uname -r)/build M=... modules, exit 0, no warnings) --
confirms the new probe loops are syntactically/semantically correct. The
full containerized boot + QEMU proof (real virtio-scsi disk -> SDA0:) runs
via the workflow_dispatch CI leg fired alongside this PR.

Refs vms-ddc, vms-47d, vms-050 (oracle-driven UX-fidelity).

Test plan

• Local vms.ko module build (host kernel headers) -- clean, no warnings
• CI workflow_dispatch run on this branch -- kmod/acceptance legs (skipped on normal PR triggers) prove the real virtio-scsi disk -> SDA0: enumeration

🤖 Generated with Claude Code

https://claude.ai/code/session_018q4jvj6GCkdbYjRvpy7Vwu