…E for Iceberg tables (#63240)

### What problem does this PR solve?

Problem Summary:

`SHOW CREATE TABLE` for Iceberg external tables omitted the `PARTITION
BY LIST (...) ()` clause entirely. The generated DDL was not
reproducible — running it as a new `CREATE TABLE` would create an
unpartitioned table, silently losing the partition strategy.

`Env.getDdlStmt()` for `ICEBERG_EXTERNAL_TABLE` reconstructed `ORDER BY`
and `LOCATION` from Iceberg metadata but had no code path for the
partition spec.
  
Added `getPartitionSpecSql()` to `IcebergExternalTable`, which reads the
live `PartitionSpec` from the Iceberg catalog and converts each
`PartitionField` back to Doris DDL syntax. Wired into `Env.getDdlStmt()`
between `ORDER BY` and `LOCATION`.

*Transform mapping* (all Iceberg standard transforms covered):
  -  isIdentity() → col
  -  isBucket(n) → BUCKET(n, col)
  -  toString() = "truncate[n]" → TRUNCATE(n, col)
  -  toString() = "year/month/day/hour" → YEAR/MONTH/DAY/HOUR(col)
  -  isVoid() → skipped (dropped partition field from evolution)

Uses `isVoid()` and `isIdentity()` from the public `Transform` interface
where available, falls back to the spec-defined canonical `toString()`
names for remaining transforms (stable across all Iceberg versions as
they are used in metadata serialisation).


Co-authored-by: Sivarajan Narayanan <narayanan_sivarajan@apple.com>