rgw/dedup: extend shard tiers, add overflow guard, and fix bucket filter

The dedup subsystem was limited to 512 MD5 shards with 10% headroom,
  capping effective operation at ~4 billion objects.
  - Raise MAX_MD5_SHARD from 512 to 2048 and add new shard tiers for
    1024 (2GB), 2048 (4GB) shards, supporting up to ~213 billion objects.
  - Increase headroom from 10% to 20% to improve hash table load factor.
  - Reject pools that exceed the maximum capacity with -EOVERFLOW
  - Apply bucket filter in collect_all_buckets_stats() so that filtered
    buckets do not inflate the object count used for shard sizing.
  - Update sizing table, comments, and documentation to reflect the new
    tiers and capacity limit.

Signed-off-by: benhanokh <gbenhano@redhat.com>