[kvlonge]

Why

Large projects repeated model change detection for optional selection diagnostics and rebuilt project-wide SQL-test indexes for every test case. That work scaled with both project and test counts even when the results were identical.

Changes

• skip stale-warning detection and identity work when selection diagnostics are disabled
• reuse execution change results only when scope identities and query-tracking semantics match
• preserve independent diagnostic detection when query tracking is disabled
• treat complete diagnostic results as authoritative during warning assembly
• share SQL-test model/function indexes, fixture model maps, and identical chain topology within one plan
• add deterministic work-count and semantic-equivalence regression coverage

Verification

• uv run pytest tests/unit/src/sqlbuild/compiler/planner/_helpers/planning/test_identities.py tests/unit/src/sqlbuild/compiler/planner/_helpers/test_selection_staleness.py tests/unit/src/sqlbuild/compiler/planner/_helpers/changes/test_detect.py tests/unit/src/sqlbuild/compiler/planner/_helpers/test_fixture_planning_work.py tests/unit/src/sqlbuild/compiler/planner/_helpers/sql_test_assembly/test_plan_test.py tests/unit/src/sqlbuild/compiler/planner/main/test_execution.py tests/integration/src/sqlbuild/compiler/planner/_helpers/test_fixture_planning.py tests/integration/src/sqlbuild/compiler/planner/main/test_build_execution_plan.py -n auto --dist loadfile
• make check-ci
• make test (6,771 passed)