[flink] Avoid creating empty lake writer in TieringSplitReader

[luoyuxia]

Purpose

Linked issue: close #3290

Fix a tiering log reader bug where TieringSplitReader creates a lake writer before it knows whether the current polled batch contains any record below the split stoppingOffset.

Brief change log

• Delay lake writer creation in TieringSplitReader.forLogRecords until the first record with record.logOffset() < stoppingOffset is encountered.
• Keep the existing split finishing path unchanged so offset advancement without actual writes no longer produces an empty lake writer commit.
• Generative AI disclosure: Codex (GPT-5).

Tests

• Ran ./mvnw -pl fluss-flink/fluss-flink-common -DskipTests spotless:check checkstyle:check.
• Did not run unit or integration tests, per request.

API and Format

No API or storage format changes.

Documentation

No documentation changes.

[fresh-borzoni]

+1
nit: just a small regression test will be great, and also I checked quickly SnapshotSplitRecords, it's safe, but relies on BoundedSplitReader logic:

• If a snapshot has zero records, BoundedSplitReader.readBatch() returns null
• if Empty iterator -> writer created, but nothing written. Next call writes records to the same writer, so it's safe

Mb we wish to add the same symmetrical fix there or just leave comment, so that it's less brittle

[beryllw]

LGTM!

[fresh-borzoni]

Nice, ty 👍