[kvlonge]

Why

sqb plan and sqb build planned every selected SQL test from scratch after warehouse inspection finished, even though test planning never touches the warehouse. On a large project this took most of the "Generated plan" phase. build --no-tests planned tests it would never run.

Changes

• SQL test planning runs on a background daemon thread started once scopes are resolved, overlapping warehouse inspection. The result is joined before plan assembly and reused only when its selected keys match the execution scope; otherwise tests are planned as before. Planning errors are raised at the same point as before.
• build --no-tests, direct and virtual, no longer plans SQL tests. As a result, an invalid test fixture no longer fails build --no-tests; normal builds still reject it.
• A failed or interrupted plan never waits for the background thread before exiting.

Verification

• Real large-project sqb plan, 3 paired runs: "Generated plan" went from 17.6–19.3s to 3.0–4.2s. The full plan was identical to main apart from live data-version fields, which also differ between two main runs.
• New real-CLI integration tests for direct and virtual builds with an invalid fixture, with and without --no-tests. Unit tests cover the background result, disabled planning, error propagation and the daemon worker.
• make test and make check-ci pass, public-repo scan clean, one review cycle with findings fixed and re-verified.