Add ZFSDDTStats for OpenZFS dedup table (DDT) stats - #878

Adds a parser for the OpenZFS dedup table (DDT) statistics in
/proc/spl/kstat/zfs/<pool>/ddt_stats_<checksum>, so node_exporter's zfs
collector can expose them (see prometheus/node_exporter#3435). This is the first
ZFS code in procfs.

What it does

FS.ZFSDDTStats() globs spl/kstat/zfs/*/ddt_stats_* and returns one
ZFSDDTStats per file:

• Pool and checksum come from the path (parent directory name, and the
  filename with ddt_stats_ trimmed). A DDT belongs to the whole pool, so
  there is no dataset.
• One file per dedup-capable checksum. ZFS creates sha256, sha512, skein,
  edonr and blake3 files for every pool, and only the checksum in use has
  non-zero values.
• Format: the files are named kstats (name type data). The three
  *_rate fields are KSTAT_DATA_UINT32; the rest are KSTAT_DATA_UINT64.
  All of them are returned as uint64.
• Unknown field names are ignored, so fields added in later ZFS versions
  don't break parsing.
• No files means an empty result, not an error. That covers OpenZFS older
  than 2.3 and systems without ZFS.

Field definitions are in ddt_kstats_template in
module/zfs/ddt.c.
The counters reset each time the pool is imported.

Reading is done with parsers.ReadFileNoStat (these files report size 0),
and parsing is a separate function, following CONTRIBUTING.md.

Testing

• Fixtures: ddt_stats_sha256 and ddt_stats_blake3 for pool1,
  captured unmodified from OpenZFS 2.3 on Debian 13.
• Fixture test: compares the full parsed result for both files, including
  the UINT32 rate fields. The sha256 values are internally consistent, e.g.
  lookup 800 = lookup_new 400 + lookup_existing 400.
• Parser test: a table test for parseZFSDDTStats covering empty input,
  a missing header, a non-numeric value, an unknown field name, an unexpected
  type, and a UINT32 row.
• make test and make lint pass.

Follow-up

Once this is released, I'll open a node_exporter PR that uses
fs.ZFSDDTStats() to add DDT metrics to the zfs collector. That adds 17
metrics per checksum file, which is 85 series per pool (5 checksums × 17
fields).

cc @SuperQ @discordianfish