tools/cephfs_mirror: Expose per-directory snap metrics via perf counters

tools/cephfs_mirror: Expose per-directory snap metrics via perf counters

Introduce a new labeled perf counter group, cephfs_mirror_directory, so
per-directory snapshot mirror progress can be scraped via "counter dump" and
exported to Prometheus by ceph-exporter (e.g.
ceph_cephfs_mirror_directory_current_sync_bytes).

Design
------

* One PerfCounters instance per mirrored directory on a peer, keyed in
  m_directory_perf_counters and registered on the daemon-wide
  PerfCountersCollection.
* Labels on each instance (flat counter dump array entries):
    - source_fscid, source_filesystem
    - peer_uuid, peer_cluster_name, peer_cluster_filesystem
    - directory (dir_root, e.g. "/parent/d1")
  The peer_uuid label disambiguates the same directory path mirrored to
  different peers.
* Counters are created in init() and add_directory(), removed in
  remove_directory() and the PeerReplayer destructor.
* Priority follows cephfs_mirror_perf_stats_prio (same as
  cephfs_mirror_peers).

Update path
-----------

Live / current_syncing_snap gauges are refreshed from
update_directory_current_sync_perf_counters(), called by
refresh_directory_current_sync_perf_counters() from the per-peer tick
thread (run_tick()). Each cephfs_mirror_tick_interval seconds (default 5)
the tick thread updates counters for each registered (actively syncing)
directory.

Counters registered (schema)
----------------------------

All of the following are added to the builder in this commit. Only the
"current sync" and dir_state fields listed under "Updated in this commit"
are written here; last_synced_snap and per-directory snap summary counters
are registered for a follow-up commit that updates them when stats change.

Directory state
  dir_state (gauge u64)
    0 = idle, 1 = syncing, 2 = failed
    Maps peer_status top-level "state" (numeric; no string values).

Current syncing snapshot (peer_status "current_syncing_snap")
  [Updated in this commit]
  current_snap_id          - snapshot id being synchronized
  current_sync_mode        - 0 = full, 1 = delta (snapdiff)
  current_read_bps         - bytes/sec read (raw, not formatted)
  current_write_bps        - bytes/sec written
  crawl_state              - 0 = N/A, 1 = in-progress, 2 = completed
  crawl_duration_seconds   - crawl duration; in-progress uses now - start
  datasync_wait_state      - 0 = none, 1 = waiting, 2 = complete
  datasync_wait_duration_seconds
  current_sync_bytes       - bytes synced so far for this snap
  current_total_bytes      - total bytes for this snap
  current_sync_bytes_percent - basis points (1745 = 17.45%)
  current_sync_files
  current_total_files
  current_sync_files_percent - basis points
  current_eta_valid        - 0 = calculating, 1 = ETA available
  current_eta_seconds      - ETA in seconds when valid

Per-directory snapshot summary (peer_status snaps_*)
  [Registered only; not updated in this commit]
  snaps_synced, snaps_deleted, snaps_renamed

Last synced snapshot (peer_status "last_synced_snap")
  [Registered only; not updated in this commit]
  last_snap_id
  last_crawl_duration_seconds
  last_datasync_wait_duration_seconds
  last_sync_duration_seconds
  last_sync_timestamp      - utime_t / seconds since epoch
  last_sync_bytes
  last_sync_files

When idle or failed, current_* counters are zeroed and dir_state reflects
0 or 2 respectively.

Fixes: https://tracker.ceph.com/issues/73457
Signed-off-by: Kotresh HR <khiremat@redhat.com>