Fix stale values in ungrouped aggregate queries with no matching rows

[vimalprakashts]

Summary

• Fix ungrouped aggregate queries returning stale non-aggregate column values instead of NULL when no rows match the WHERE clause
• Emit NullRow for all btree table and index cursors in the empty-result path
• Add null-flag check to IdxRowId so it returns NULL (matching Column behavior)

Bug

CREATE TABLE t(x INTEGER, y TEXT);
INSERT INTO t VALUES(1, 'a'), (2, 'b');
SELECT x, COUNT(*) FROM t WHERE 0;

Expected (SQLite): NULL|0
Actual (Turso before fix): 2|0 — leaks the last scanned row's value

The same issue affects index-only scans (IdxRowId returning a stale rowid).

Root Cause

The ungrouped aggregation codegen has a "no rows" fallback path that evaluates non-aggregate columns when the scan loop never executed. This path nulled coroutine output registers (for CTEs/subqueries) but did not issue NullRow on btree/index cursors. The Column instruction already checks the null flag, but IdxRowId did not — so index-based plans also leaked stale rowids.

Fix

1. core/translate/aggregation.rs: In the no-rows fallback, emit NullRow for every btree table cursor and index cursor (looked up via resolve_cursor_id_safe to handle covering-index cases where the table cursor isn't allocated).
2. core/vdbe/execute.rs: Add a null-flag check at the top of op_idx_row_id — if the cursor is in NullRow state, return NULL immediately instead of reading a stale rowid.
3. core/types.rs: Add Cursor::get_null_flag() accessor.

Test Plan

• New sqltest: testing/runner/tests/ungrouped-aggregate-nullrow.sqltest — covers table scans, index scans, multi-table joins, CTEs, WHERE-false and WHERE-no-match variants
• make -C testing/runner run-rust passes
• cargo test passes
• cargo clippy --workspace --all-features --all-targets -- --deny=warnings passes

[turso-bot]

Please review @PThorpe92