[SEA-NodeJS] SEA async execute, TLS & connection/statement options

[msrathore-db]

Recreation of #413 pointing at main (the original #413 was based on the #412 branch and got merged there instead of into main, so its changes never landed on main). Same content, replayed cleanly on top of the merged #412 (94d63bd) — zero conflicts (the #413 fork point is byte-identical to main).

What's here (the 6 #413 commits)

• SEA connection & statement options — TLS (checkServerCertificate secure-by-default + customCaCert), maxConnections, rowLimit, statementConf, queryTags.
• SEA async execute — submitStatement + poll-to-terminal (waitUntilReady) + awaitResult; cancel stays responsive.
• Review fixes (from [SEA-NodeJS] SEA async execute, TLS & connection/statement options #413's reviews, all carried over):
  • Server-driven Cancelled/Closed/Unknown throw OperationStateError (not bare HiveDriverError) so the facade mirrors cancelled/closed.
  • queryTimeout enforced client-side (poll deadline → OperationStateError(Timeout) + best-effort cancel); not forwarded to the kernel-ignored submit field.
  • customCaCert PEM check requires BEGIN+END; SeaAuth reads options via InternalConnectionOptions; corrected poll-loop comment.
  • TIMESTAMP_NTZ bound natively, TIMESTAMP_LTZ mapped to TIMESTAMP (Spark has no distinct LTZ type) — valid on both backends.
  • queryTimeout coerced via numberToInt64(...).toNumber() (handles Int64; Number(int64) was NaN).

Verification

• tsc + lint + prettier clean; SEA unit suite 231 passing; the review fixes were validated end-to-end against a live warehouse on the original [SEA-NodeJS] SEA async execute, TLS & connection/statement options #413.

Note: the kernel-rev pin (#414) and the SEA→kernel rename (#415) are stacked after this; once this merges to main I'll re-point them at main too.

This pull request and its description were written by Isaac.

[gopalldb]

🟡 P1

1. seaCancel resets isCancelled to false on kernel-cancel failure — loses user's cancel intent (SeaOperationLifecycle.ts:134-141). Brief true window can be observed by the poll loop and throw
   OperationStateError(Canceled), then the rollback hides the intent from any other consumer. Worst case: kernel-side statement keeps running while the user believes it was canceled. Fix: don't reset
   isCancelled on cancel-RPC failure.
2. Cancel during in-flight status() poll is unbounded (SeaOperationBackend.ts:372-419). Cancel arrives → flag flipped → but observation deferred until current status() resolves. No AbortSignal threaded to
   napi. Hung warehouse = unbounded latency to observe cancel. Fix: race in-flight status() against a lifecycle promise on cancel. Add a test where cancel fires while status() is mid-flight (existing test only
   covers cancel-before-poll-starts).
3. PEM marker check is order-insensitive substring match (SeaAuth.ts:201-209). "...-----END...-----BEGIN..." (e.g. proxy intercept page with both literals) passes. Fix: ordered regex /-----BEGIN
   CERTIFICATE-----[\s\S]+?-----END CERTIFICATE-----/.
4. customCaCert honored regardless of checkServerCertificate=false — silent insecure combo (InternalConnectionOptions.ts:40, SeaAuth.ts:184-223). User providing both gets verification fully off but with a
   wasted custom CA install — the secure-looking customCaCert masks the disabled verification. Fix: error or warn when both set; today's doc explicitly endorses the ambiguous combo.
5. No close() cleanup on poll-loop terminal error — kernel statement leaks (SeaOperationBackend.ts:394-415). Failed/Cancelled/Closed/Unknown/Timeout all throw without closing the kernel-side statement.
   fetchChunk does best-effort close on error; the poll loop doesn't. Fix: best-effort seaClose on all terminal-error branches.
6. rowLimit/statementConf added to cross-backend ExecuteStatementOptions but silently dropped by Thrift (IDBSQLSession.ts:30-42). Exactly the anti-pattern the PR's own code comments warn against ("silent
   drop is the worst failure mode"). User sets rowLimit: 100 on Thrift, gets full result set, no warning. Fix: warn at the Thrift session when these are passed, or move to an SEA-specific options type.

🟢 P2

• OperationStateError messages lack statement_id and state name (matches Thrift behavior, just an enhancement).
• "Secure-by-default" should be owned by the JS layer, not inherited from the kernel default. Add explicit checkServerCertificate: true when undefined for defense in depth.
• Missing test: cancel-mid-status() (the actual race the P1 Setup E2E tests with GitHub Actions #2 fix needs).
• Missing test: server-Cancelled + client-Timeout race on same tick.
• Missing test: PEM with reversed marker order, BEGIN-only, END-only.
• hasResultSet() hardcoded true even for DDL/DML — leave for now; revisit once napi exposes the kernel's has_result_set bit.