[rust] Wire batch statistics into the log batch write and read paths

[leekeiabstraction]

Summary

• Wires the V1 batch statistics collection ([rust] Add V1 batch statistics collection #4011) into the write and read paths. Closes [rust] Batch statistics read wiring #4009, closes [rust] Batch statistics write wiring #4010.
• Write: for tables with table.statistics.columns set, the accumulator hands the resolved column mapping to the Arrow builder, which emits a V1 batch with the serialized statistics placed between the header and the records data, mirroring Java's MemoryLogRecordsArrowBuilder. A serialization failure degrades to an empty statistics section, and the batch size estimate reserves room for the statistics.
• Read: the records data is located per the batch's magic, mirroring Java's DefaultLogRecordBatch.recordsDataOffset, and every decode entry point rejects unknown magic versions before the empty-batch short-circuit. This fixes the Rust reader misparsing V1 batches that Java clients and the server's projection path already produce.

Test Plan

• A cross-language golden fixture (fluss-common/src/test/resources/encoding/statistics_block.hex, generated and asserted by a new Java test) locks the Rust statistics block byte-for-byte against Java's LogRecordBatchStatisticsWriter across all supported types, nulls, both decimal widths, and both timestamp precisions. CHAR is excluded because Java's collector records no CHAR bounds.
• New unit tests cover V1 encode/decode round trips from both builder modes, empty and corrupt statistics sections, unsupported magic (including the V2 silent-drop case), and the size reserve. cargo test -p fluss-rs --lib passes 700 tests; clippy and fmt are clean.
• The end-to-end filter pushdown integration test is left to [rust] Filter pushdown integration test #3845.

🤖 AI-assisted changes - reviewed by human developer