chore(c/validation): make TestSqlPrepareUpdate{,Stream} readback order deterministic

[fornwall]

Defect

StatementTest::TestSqlPrepareUpdate and StatementTest::TestSqlPrepareUpdateStream read back the bulk_ingest table with a plain SELECT * FROM bulk_ingest — no ORDER BY — and then assert the returned rows positionally via CompareArray ({42, -42, NULL, 42, -42, NULL} and {1, 2, NULL, 3, NULL, 3} respectively), i.e. they assert insertion order.

SQL gives no ordering guarantee without ORDER BY, so a positional assertion on an unordered SELECT relies on engine-specific behavior. SQLite and PostgreSQL happen to return insertion order, which is why these pass upstream; engines that return rows in primary-key order fail the assertion (for Spanner, whose create-mode bulk ingest keys the table on a synthetic adbc_ingest_key STRING(36) DEFAULT (GENERATE_UUID()) column, the readback order is effectively random per run). The suite itself already treats this as a solved problem elsewhere: TestSqlIngestType reads back with ORDER BY "col" ASC NULLS FIRST and TestSqlBind with ORDER BY col1 ASC NULLS FIRST.

Fix

Give both readbacks the suite's existing deterministic-readback convention — ORDER BY <quirks()->QuoteIdentifier(col)> ASC NULLS FIRST — and sort the expected vectors accordingly:

• TestSqlPrepareUpdate (int64s): {NULL, NULL, -42, -42, 42, 42}
• TestSqlPrepareUpdateStream (ints): {NULL, NULL, 1, 2, 3, 3}

The RewriteSql routing and query ids from #28 are unchanged — only the default SQL string passed to RewriteSql gains the ORDER BY clause, so drivers whose dialect rejects NULLS FIRST (GoogleSQL) can still substitute an equivalent ordering via the quirk.

Verification

Verified end-to-end against the Spanner ADBC driver's validation harness (scripts/run-adbc-validation.sh in fornwall/spanner-adbc, pinned to this commit, with the two cases removed from its expected-failure list and its RewriteSql quirk substituting ORDER BY ... ASC for the NULLS FIRST form): both SpannerStatementTest.SqlPrepareUpdate and SpannerStatementTest.SqlPrepareUpdateStream pass — 84 passed / 7 skipped of 91, expected-failure gate green.

Stacks on #28 (validation-rewrite-sql-sites), which this PR is based on so the diff shows only this fix. Like #28, this is an upstreaming candidate for apache/arrow-adbc.

🤖 Generated with Claude Code

https://claude.ai/code/session_011hsYQVXuHhnuiBD7XTVo65