[FLINK-40740][pipeline-connector/fluss] Prevent duplicate split initialization during discovery

[fxbing]

What is the purpose of this pull request?

Prevent periodic Fluss source discovery from initializing and assigning the same bucket splits more than once when a previous initialization callback has not completed. Reassigning an earliest-offset split can rewind an active reader and produce duplicate records. See FLINK-40740.

Brief change log

• Track initializing physical table paths before scheduling asynchronous split creation, and release only the current batch in its coordinator callback.
• Preserve initialization failures and the existing checkpoint format; unfinished initialization remains discoverable after recovery.
• Add deterministic regressions for overlapping discovery, complete bucket assignment, independent partitions, pending readers, initialization failures, and recovery during initialization.

Verifying this change

• The overlapping-discovery regression fails against the original implementation: three buckets produce nine split assignments instead of three.
• FlussSourceEnumeratorTest (16 tests) and FlussSourceEnumStateSerializerTest (1 test) pass on both Java 11 / Flink 1.20.3 and Java 17 / Flink 2.2.0, with no failures, errors, or skipped tests.
• Spotless and Checkstyle pass. Full pipeline/CLI end-to-end tests were not run.

Documentation

• Does this pull request introduce a new feature? No.
• If yes, how is the feature documented? Not applicable.

Was generative AI tooling used to co-author this PR?

• Yes — Codex CLI; Claude and GLM assisted with analysis, implementation, and review.

Generated-by: Codex CLI 0.155.1