feat(arrow): add RecordBatch bulk insert via C Data Interface

[austin-barrington]

Add high-throughput Arrow ingest for RecordBatch and RecordBatchReader, with automatic fallback between direct libchdb FFI (chdb_insert_arrow_*) and a register → INSERT … SELECT FROM ArrowStream('name') → unregister SQL path when the symbols are absent at link time.

Breaking change: ArrowStream, ArrowSchema, and ArrowArray now wrap Arrow C Data Interface pointers (FFI_ArrowArrayStream/Schema/Array). Registered names must be queried via ArrowStream('name'), not as bare table identifiers. Add arrow_stream_table_sql() to build the correct table expression.

Also includes named column lists on the SQL insert path (maps by column name, not position), InsertOptions tuning, session helpers, example 08, integration tests, and docs. Bumps crate version to 1.4.0.

Note

Add Arrow RecordBatch bulk insert to Session via C Data Interface

• Adds insert_record_batch, insert_record_batches, and insert_record_batch_reader methods to Session and Connection, using Arrow's FFI (C Data Interface) to insert in-memory Arrow data into chDB tables without writing to disk.
• Two insert paths are supported: a registration-based SQL path using the ArrowStream('name') table function (always available), and a direct FFI path via chdb_insert_arrow_array/chdb_insert_arrow_stream (enabled when the detected libchdb exports these symbols, gated by cfg(direct_arrow_insert)).
• Adds InsertOptions in src/arrow_options.rs to configure ClickHouse insert settings (e.g. max_threads, max_insert_block_size) passed as a SQL SETTINGS clause or direct FFI options.
• The arrow feature is enabled by default (pulling in Arrow 59), so downstream crates that set default-features = false will lose Arrow support.
• Behavioral Change: package version bumps from 1.3.1 to 1.4.0 and the arrow crate becomes a default dependency, increasing compile times and binary size for all users who do not opt out.

^{Macroscope summarized ee60642.}

[CLAassistant]

All committers have signed the CLA.

[wudidapaopao]

Thanks, this is great. A few non-blocking notes, since arrow is now public API:

• Bump arrow to 59 — pinning 54 forces all downstreams onto 54 to pass a RecordBatch in.

• pub use arrow; so users go through chdb_rust::arrow::… instead of pulling their own (mismatched) copy.

• Make it optional, default-on (optional = true + default = ["arrow"]) so SQL-only users don't build all of arrow.

[austin-barrington]

@wudidapaopao Fixed, please take another look 😄

[wudidapaopao]

LGTM! Just a minor cargo fmt CI failure to fix.