blockdev: Fall back to blkid -p when udev is unavailable

[cgwalters]

See osbuild/osbuild#2428

When running inside a container or sandbox without /run/udev bind-mounted (e.g. osbuild's bwrap), lsblk returns null for partition metadata fields like parttype (partition type GUID) and pttype (partition table type), because these come from the udev database.

This causes partition discovery to fail -- most visibly on ppc64le where the PReP partition can't be found by GUID, but also affecting ESP discovery on all architectures. On x86_64 UEFI, bootupd silently falls back to mounted-ESP detection, masking the problem; on ppc64le there is no fallback and the install fails hard.

Add a blkid -p fallback in backfill_missing() that probes partition metadata directly from the disk when udev is absent. The udev check tests for /run/udev/data (the actual database directory) rather than /run/udev, because bootc's ensure_mirrored_host_mount() creates an empty /run/udev directory on the bwrap tmpfs that is_same_as_host() then considers 'already mounted'.

Tested end-to-end via the BIB tmt test (plan-33-bib-build) with a patched bootupd that removes the mounted-ESP fallback entirely.

This follows the same approach coreos-installer uses (PRs #1511, sfdisk-based workaround originally added in PR #688.

Assisted-by: OpenCode (Claude Opus 4)

[gemini-code-assist]

Code Review

This pull request implements a fallback mechanism for block device metadata discovery when the udev database is unavailable, such as within container sandboxes. It introduces blkid -p as a secondary source for partition and table types, updates the Device backfilling logic, and adds new internal CLI subcommands for block device inspection. Feedback was provided to optimize the backfill_missing function by caching the device path to avoid redundant string operations.