Testing: Component binaries (T-07) #341

Part of the test strategy (captain, 2026-10-05: scenarios that test the changes together; cleaner, faster suites; TESTING.md). Epic #305.

Scope. Per layer, one binary with a harness module: tests/storage/ (storage_engine, persist_roundtrip, persist_crash, wal_transaction_recovery, crash_recovery, restart_equivalence, data_dir_lock, data_dir_loss, online_checkpoint, backup_safety, backup_restore, name_validation), tests/handler/ (client_handler, session_schema, input_validation, statement_status, statement_counts, atomic_program, fact_program, pinned_proof, why_provenance, why_not, hnsw_index, size_limit, memory_limit, authz_multistatement, server_startup, bootstrap_secret, multi_clause_rule, mutual_recursion, wildcard_rule, string_literal, comparison, expression_comparison, typed_arithmetic, aggregate_correctness, result_limit, snapshot_sharing, optimizer_config, magic_sets_regressions, fastpath_regressions, the surviving error_handling cases), tests/ws/ (ws_session minus its config assertions, ws_pipeline, ws_cancel, ws_delivery, ws_expect_revision, ws_hygiene, notification_cursor, credential_revocation, auth_hardening) and tests/standing_query/ (unchanged; its parameterized/sharing modules are frozen until V11 deletes them). Delete integration_test.rs, advanced_integration_tests.rs, example_execution_tests.rs, flights_reachability_tests.rs after confirming each case has a spec (list in the PR); keep example_verification.rs as a unit test of the spec corpus. nesting_depth_tests stays its own binary (process-wide limit) until the limit is per handler. Split per layer: four PRs.

Acceptance. Test binary count from 73 to about 12; every deleted test named with its replacement; the component tier runs in under 3 min on 4 cores with the per-binary timing table attached; no behaviour change.