fix: preserve row mode in native pipelines

[zfaustk]

What changed

Native pipeline entries now retain each query's array mode, and pg-native uses that mode when it builds the corresponding pipeline result.

The regression test mixes rowMode: 'array' and the default object mode in one connected pipeline batch, so it also verifies that row mode does not leak between queries.

Why

The non-pipeline native path applies NativeQuery._arrayMode before reading a result, but the pipeline path dropped that per-query value. A rowMode: 'array' query therefore returned object rows when pipeline: true.

This is observable downstream in drizzle-team/drizzle-orm#6118: interpreted mapping produced undefined fields and JIT mapping threw TypeError: object is not iterable.

Verification

• New integration test fails on ff9d775 in native mode (object row returned instead of [10]).
• The focused pipeline integration file passes in both JavaScript and native modes after the fix.
• The original Drizzle reproduction passes for native direct queries plus interpreted and JIT mapping, with and without pipelining.
• Repository-wide ESLint passes; the three changed files pass Prettier check.
• The pg unit suite passes.

The canonical full yarn test matrix was not completed locally: its Yarn 1 workspace install caches unrelated multi-platform optional binaries beyond this executor's bounded disk budget. GitHub CI remains the authoritative full matrix.

Compatibility

No public API is added or changed. The native pipeline path now matches the existing per-query rowMode behavior of the non-pipeline and JavaScript clients.

AI assistance

Codex was used for investigation, implementation, and test execution. No human-review attestation is made.

[brianc]

ohhhh very nice! good find 👍