[iceberg] Fix Iceberg metadata after a rollback

[plusplusjiajia]

Purpose

After a rollback, later commits reuse the deleted snapshot ids, but IcebergCommitCallback finds the old vN.metadata.json and returns early — Iceberg readers keep seeing the pre-rollback rows forever (live rows (1,10),(3,30) read back as (1,10),(2,20)). The schema story also goes wrong: snapshot entries are stamped
with the latest schema id, and reused schema ids stay bound to abandoned definitions.

The fix:

• every snapshot summary records a commit identity (the Paimon snapshot's uuid; legacy metadata without it stays trusted); existing metadata and bases are kept only on an identity match, otherwise rebuilt — inheriting the replaced metadata's table uuid and last-column-id high-water mark;
• the abandoned suffix is retired from the actual files, only while the snapshot is still the Paimon head, after the replacement and catalog pointer are durable; retries recommit the catalog pointer with the previous version as base, and the REST committer treats an already-current catalog as a no-op instead of recreating
  the table;
• snapshot entries record their own schema id; reused schema ids are detected by content comparison and trigger a rebuild; a plain schema-pointer rollback keeps history and refs.