Fix main-red witness imports and located floor terminal refusals

[gunbai-bot]

Summary

• Root cause (run 31267988397 / 8ef88e95ee): batch 3 reported 25 discovery witness reds (no such function: …) in extdeps_anemia_grounding_witness_test.dag, extdeps_round2_grounding_witness_test.dag, and hardware_selection_witness_test.dag. Those files only imported ampere and relied on pool-membership resolution for every other symbol; once the closure no longer dragged those modules in, resolution failed. This PR adds explicit import rows for each authority (sql/http/cloud/gpu/cpu/memory/product/hardware_selection).
• Loudness gap: the worker emitted [falsifier-failure-class] with the full batch detail, but floor_terminal_fast_exit wrote only walk terminal exit code 1 into the terminal receipt, which is what the coordinator surfaced. walk_terminal_detail now projects mode, ci_failure_class_arm, and joined failure_details into the receipt so a terminal exit 1 names the refusal.

Not in scope (escalation): run 31269330339 (e52f99fac1e, #8046) completed the discovery corpus (drain receipt + measurement at ~45m) but the CI step failed after ~71m with no ✗ FAIL / walk terminal line in the failed log — consistent with step timeout or outer cgroup kill after the receipt, not witness reds. That belongs to the run-stability / step-budget lane.

Test plan

• cargo test -p v1-compiler --bin claim_executor walk_terminal
• claim_batch --function w_sql_dialect_postgres_label on extdeps_round2_grounding_witness_test.dag
• claim_batch --function w_cpu_tdp_watts_power_dimension on extdeps_anemia_grounding_witness_test.dag
• claim_batch --function downclock_is_rank_dependent on hardware_selection_witness_test.dag
• CI floor on PR

Made with Cursor