[ShawnChen-Sirius]

Summary

• Accept schema-bearing Arrow Tables with zero batches as zero-row query sources and inserts.
• Add AbortSignal support to direct Arrow queries, output readers, input readers, and reader inserts; preserve native teardown ordering and document early-settle behavior.
• Add regression coverage for empty sources and cancellation before the first batch, during fetch, and during insert.

Validation

• npm run build:ts
• 75 Arrow-focused tests passed.
• Full test:v3 run: 559 passed, 11 skipped, one localhost watchdog test failed because the sandbox denied listening on 127.0.0.1 (EPERM). That test passed when rerun outside the sandbox (26/26 in its file).

Arrow IPC byte-stream input is intentionally outside this PR; direct input still accepts Arrow objects and batch iterables.

Note

Add AbortSignal support and zero-batch Table sources to Node Arrow APIs

• Session.queryArrow, queryArrowStream, registerArrowReader, insertArrow, and insertArrowReader accept an optional AbortSignal. Abort is checked before native work starts, while awaiting batches, and after query resolution; aborts settle the caller early with ChdbAbortError while pending native work drains first.
• On abort, readers are closed, the native reader is unregistered, and upstream iterator cleanup is requested via return() without waiting for a blocked next().
• registerArrowTable and insertArrow accept a Table with a schema but zero batches as a zero-row source; emptyArrowBatch builds the structural zero-row batch and rejects an empty schema.
• DirectArrowOptions in index.d.ts gains the signal property, and arrow.md documents the new behaviors.
• Risk: aborting a buffered query ends the caller's wait but native computation continues until teardown; there is no rollback guarantee for inserts. Check withArrowAbort and ChdbArrowReader in index.js.

^{Macroscope summarized 14ca7fa.}

[ShawnChen-Sirius]

@wudidapaopao, please review this PR

[wudidapaopao]

LGTM