[flink] Fix limit pushdown for lake union read

[QuakeWang]

Purpose

LIMIT pushdown previously always used the Fluss-only limit scan path. For datalake-enabled FULL scans, this skipped the lake source, so union reads could miss records that had already been tiered to the lake.

Brief change log

• Avoid the Fluss-only limit scan path when a FULL scan needs to read from the lake.
• Let the normal Flink source path perform lake + Fluss union reads in that case.
• Add a Paimon IT covering LIMIT reads from an expired Fluss partition whose data still exists in Paimon.

Tests

• ./mvnw -pl fluss-flink/fluss-flink-common,fluss-lake/fluss-lake-paimon -am -DskipTests -DfailIfNoTests=false test-compile

API and Format

Documentation