Preserve row-cap metadata through binary query APIs #9

Priority

p0 — Known capped MessagePack results are represented without their cap signals.

Gap

PR #2 decoder explicitly reconstructs QueryResponse without rows_capped and row_cap, although both server baselines emit them. PR #6 adds model fields/defaults but leaves that decoder unchanged. This issue covers the residual binary transport gap, not duplicate JSON model work.

Decoder, producer. Unchanged decoder/constructor fixture confirms omitted keys; full MessagePack transport remains untested. Arrow exposes cap information in trailers, requiring separate transport investigation.

Acceptance

• Extend/reuse feat(sdk): v0.2.0 — feature parity with Arc server + hardening #2/feat(query): expose truncation and row-cap metadata #6 rather than duplicate their implemented model changes.
• Preserve present/absent MessagePack row-cap fields through public sync/async query calls.
• Test JSON separately and keep failure truncation distinct from policy row caps.
• Prove Arrow trailer availability and decide a backwards-compatible metadata interface for Arrow/pandas/polars; document any remaining limit.
• Preserve integers/nulls/timestamps across wire formats; test nanosecond precision explicitly.

Tracking: https://github.com/Basekick-Labs/ops/issues/40. Analysis baselines: SDK main 46d3b66 and pending PRs #2/#3/#5/#6; Arc release e83bb1e (v26.09.3) and main a63f4a8. Implementation is unscheduled; the parent analysis deadline is 2026-10-30. No owner or delivery commitment is implied.

Verification must record exact versions/commands/results and separate Manual Test from IA Test. Require sync/async cases, negative authorization/validation cases, and existing-regression preservation. Local Docker is appropriate for end-to-end server checks; inventory and clean up only test resources.