vm-repair scripts: fix resource-disk exclusion and add Linux NVMe readiness detection

[Edwin Bernal Microsoft (EdwinBernal1)]

Summary

Fixes Azure resource-disk exclusion in the NVMe-aware Windows disk-discovery helper and adds a read-only Linux detector for SCSI-to-NVMe boot readiness.

The resource disk is reported as SAS on current Azure repair VMs, not the literal SCSI, and PowerShell 5.1 returns a scalar for a single partition. The old check therefore treated the temporary resource disk as a repair target. The fix recognizes the SCSI-family bus types and array-wraps partition results before counting them.

The new linux-detect-nvme-readiness run-id mounts the attached root filesystem read-only and reports unstable device references, NVMe-driver presence in the newest initramfs, dracut host-only configuration, distro identity, attached-disk transports, and one machine-readable [NVME-EVIDENCE-JSON] record. It does not modify the attached OS disk.

Requested by / source

• Feature 37971694
• Requested by Bila / Gabriela Limoli
• Companion planning: NVMe-PR-Scripts.md

Type

• Bug fix
• Feature
• Breaking change
• Documentation only

Changes

• src/windows/common/helpers/Get-Disk-Partitions-v3.ps1
  • Recognizes SCSI, SAS, and RAID as the SCSI family when excluding the Azure resource disk.
  • Array-wraps Get-Partition output before counting it.
  • Reports the correct partition count in the warning path.
• tests/test-get-disk-partitions-v3.ps1
  • Adds a standalone regression test for a SAS resource disk represented by a scalar partition result.
  • Confirms the attached OS disk remains a repair target.
• src/linux/linux-detect-nvme-readiness.sh
  • Adds read-only Linux NVMe-readiness detection and structured evidence output.
  • Guards expected nonzero commands for the run driver's bash -e behavior.
  • Ensures mounted filesystems are cleaned up on exit.
  • Escapes guest-controlled values before writing evidence JSON.
• map.json
  • Registers the new permanent run-id linux-detect-nvme-readiness.

Version & changelog

• Not applicable: this PR changes repair-script-library, not the versioned Azure CLI extension package.

Testing

Completed locally

• Branch created from latest origin/main (0df7fa7, merge of PR vm-repair scripts: add read-only win-detect-nvme-readiness detection script #144)
• Regression test fails against pre-fix origin/main: resource disk is returned and the test finds two targets
• Regression test passes with this fix: SAS resource disk is excluded and OS partition F: remains
• git diff --check origin/main...HEAD
• Git Bash syntax check: bash -n src/linux/linux-detect-nvme-readiness.sh
• map.json parses
• All 28 run-ids are unique
• Every mapped script path exists
• PSScriptAnalyzer — unavailable in the current environment

Cross-repo impact

• Adds one public repair-script run-id: linux-detect-nvme-readiness.
• No Azure CLI extension code change is required for run-id discovery because list-scripts reads map.json.
• Companion extension work is tracked separately in NVMe-PR-Extension.md.
• Decision on overlap with repair-script-library PR Add the offline repair foundation helpers (1 of 3) #143 remains to be recorded before R1 implementation begins.

Security review

• The Linux detector mounts the attached root filesystem with -o ro and contains no write operation against that filesystem.
• Cleanup is registered with an exit trap so unexpected failures do not intentionally leave the root filesystem mounted.
• Guest-controlled evidence values are JSON-escaped before serialization.
• Device overrides and discovered paths are quoted at command boundaries.
• No eval, bash -c, sh -c, command construction from untrusted input, credential logging, secrets, or customer data were introduced.

Backward compatibility

• Existing run-ids and scripts are unchanged.
• Existing v1/v2 disk helpers and their callers are untouched.
• The v3 fix narrows results only by correctly removing the documented Azure temporary resource disk.
• linux-detect-nvme-readiness is additive and read-only.