Implement the ldiskfs block device scanner using libext2fs to achieve high-performance MDT and OST object scanning.

Generate an LFU Object Stream by reading MDT (and OST) metadata directly from the block device in userspace, via libext2fs, at device read bandwidth. This replaces client-side lfs find traversal for the initial rollout.

It follows e2scan (in Lustre e2fsprogs history) and Lester, the Lustre Lister. It runs independently of the MDS/OSS service and needs only read-only access to the underlying block device.

In scope

• Full-device enumeration of ldiskfs inodes on an MDT or OST
• FID recovery and object classification from Lustre xattrs
• Attribute extraction into an Object Stream, driven by a caller-supplied mask
• Pushed-down filter evaluation, ordered by I/O cost
• Parallel scan within one device; checkpoint and restart

Out of scope

• Repair of any kind. LFU reads; LFSCK and OI Scrub repair. The scanner has no write path and opens the device O_RDONLY.
• Pathname construction — an Output Format concern (Path resolution priced separately), though this module emits the linkea bytes that make it possible.
• Cross-target merge — a Filter Rule concern, server-side
• ZFS and WBCFS backends — those are served by the OSD API Input Scanner, which is why that module exists.

Why this module first

It requires no kernel changes and no Lustre server modification, so it works against unmodified existing servers. That is what makes the "no flag day" rollout real.