[iceberg] Retry transient REST DNS failures

[arthurgaubil]

Purpose

Retry Iceberg REST catalog DNS lookup failures with configurable exponential backoff. Five retries with a one-second initial delay are used by default; setting metadata.iceberg.unknown-host-retry.max-retries to 0 preserves immediate failure. Exhausted retries still propagate so publication cleanup and durable recovery remain safe.

Tests

mvn -o -Ppaimon-iceberg,fast-build -pl paimon-iceberg -DwildcardSuites=none -Dtest=IcebergRestMetadataCommitterTest test

All 30 tests passed, including transient recovery, exponential delay, disabled retries, exhaustion, and non-DNS failures. Generated configuration documentation was refreshed with mvn -o package -Pgenerate-docs -pl paimon-docs -DskipTests.

[JingsongLi]

Retrying a transient DNS failure on the commit path is reasonable, and the cause-chain walk correctly handles wrapped exceptions.

One note for the record: with the defaults (max-retries 5, initial delay 1 s, doubling), a genuinely dead host pins the commit thread for ~31 s (1+2+4+8+16) before failing. That is a reasonable trade-off and both knobs are configurable, but worth spelling out in the option docs so commit latency is not surprising. Also, this intentionally only covers UnknownHostException and not other transient connection errors — fine as a narrow scope, just worth documenting.