[connect/tidb] Split key ranges by TiKV region via FLIP-27 enumerator

[lifulong]

What is the purpose of this pull request?

TiDB CDC used to split a table by evenly partitioning the entire long handle space (Long.MIN_VALUE ~ Long.MAX_VALUE).
For clustered AUTO_INCREMENT primary keys, real handles only occupy a tiny slice of that space. With parallelism 8, almost all rows land on a single subtask, which then OOMs during KVClient.scan.

This change assigns each reader a contiguous chunk of real TiKV regions covering the table, so snapshot and incremental CDC scale with region count instead of the theoretical long domain.

Region topology is discovered once by a FLIP-27 enumerator. Each parallel SourceFunction.open() used to query PD independently; if region split/merge happened while tasks started minutes apart, assigned key ranges could overlap (duplicate CDC) or leave gaps (lost data). The enumerator snapshots region boundaries, assigns one contiguous KeyRange per reader, and restore reuses checkpointed splits instead of rediscovering PD.

Runtime region split/merge is still handled by CDCClient inside the assigned key range. That only remaps physical regions onto a frozen key-space ownership and does not re-partition work across readers.

Brief change log

• Add a thin FLIP-27 TiKVSource (TiKVSourceEnumerator + TiKVSourceReader)
• Enumerator queries PD once via RangeSplitter.splitRangeByRegion, clips/sorts/merges contiguous region chunks, and assigns one TiKVKeyRangeSplit per subtask
• Reader scans snapshot then streams CDC only for its assigned key range; resolvedTs is stored on the split
• Enumerator checkpoint stores unassigned splits; restore never rediscovers regions
• Fail the job if region meta cannot be fetched (no silent fallback to task0-owns-all-table)
• Table API uses SourceProvider; DataStream TiDBSource.builder().build() now returns TiKVSource (use env.fromSource)
• Deprecate TiKVRichParallelSourceFunction / buildSourceFunction()
• Keep the old handle-space helpers as deprecated APIs
• Update DataStream docs (en/zh)

Verifying this change

• Added/Updated unit tests in TableKeyRangeUtilsTest and TiKVKeyRangeSplitSerializerTest
• Manually tested by our produce env job (region-based split)

Documentation

• Does this pull request introduce a new feature? no (bugfix / correctness of parallel split)
• If yes, how is the feature documented? docs / JavaDocs (DataStream example switched to fromSource)

---

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

• Yes (Cursor)

[lifulong]

@leonardBang @GOODBOY008 @lvyanquan can you help review this change?