LFU starts client-side so the record contents and the consumer API are settled before any kernel or wire-protocol work is committed to.

Most of the crawler is already in tree: LU-17814 landed `liblustreapi_pfind.c` — threaded traversal, attributes via`LL_IOC_MDC_GETINFO_V2`, `statx()` fallback — and `llapi_find_with_cb()` /`llapi_find_cb_t` are exported. This is not a new crawler, but a callback thatfills a per-object record instead of printing.

New `lustre/utils/liblustreapi_scan.c`:

struct llapi_scan_rec;                  /* one object: FID, attrs, layout */
typedef int (*llapi_scan_cb_t)(const struct llapi_scan_rec *rec, void *data);
int llapi_scan_namespace(const char *path, struct llapi_scan_param *param,
                         llapi_scan_cb_t cb, void *data);

Exported by the existing `llapi_*` rule in `liblustreapi.map`; man page in `Documentation/man3/`. Named for what it does rather than for the programme, since renaming an exported symbol later is an ABI event.

Its own ticket because the scanner has several queued consumers — `lfs find` first, then the PCC-RO and Trash Can cleanup tools, FLR/EC resync, HSM — and each should be landable on its own.

Scope: this defines an in-memory record and a callback contract, not a serialization. The Object Stream format is still open between FlatBuffers and Cap'n Proto, so nothing here should freeze it.

Review note: pushed as one stacked Gerrit series with the `lfs find`
conversion, so the API's first caller is visible in the same push.