Main-red hotfix: stale roster row from #7060 + stale script-module import orphaned by #7064

[gunbai-bot]

Main ci has been red since ~16:33Z. This PR fixes FOUR independent latent breaks behind it — all cross-lane merge artifacts from today's batch, surfaced layer-by-layer because the floor stops at the first failing batch (each fix exposed the next).

Fix 1 — phantom witness (batch 2, every main push red). #7060 moved scanner_clean_tree_meta_exec_holds_on_rostered_corpus to the long lane and deleted the fn from the old entry file, but the explicit CommitWitnessClaim roster row in dag/gunbc/commit_workflow.dag still enrolled it ('no main function found', e.g. main run 29940402396). Fix: roster row keeps only the fixture-scan witness still living in the entry file.

Fix 2 — unresolved import (batch 1 on wide-closure PRs). #7000 added host_runner_memory_cap_verify consuming fleet_show_effective_read_script; #7064 (branch predating #7000) deleted that module. Latent on main (push runs skip compile-clean on empty diff observation). Fix: the three memory-prop reads route through the typed systemctl_show_property_read_local seam (slice1 precedent), constructing RunnerUnitLiveReadRow → runner_unit_live_read_typed — the local-transport mirror of fleet_runner_unit_memory_live_read. Fail-closed behavior preserved.

Fix 3 — generated-artifact drift (batch 4, latent behind batch 2). #7071 hand-added an in-flight snapshot paragraph directly to generated ROADMAP.md (in no authority carrier — silent-ibex-417 notified to re-land it authority-side if wanted), and ci.yml carried a stale trailing blank line. Fix: both files regenerated byte-identical to generated_artifact_gate main_wet output, run locally by execution.

Fix 4 — layering gate cannot resolve (batch 4, latent behind batch 2). The gate transport's 13-source scoped run red with 'unresolved type LayerPrefix' — layering_imports.dag referenced it bare and nothing in the scoped closure pulls v2.std.layer (whole-tree pool masks this everywhere else). Fix: explicit import completes the closure by construction; the gate now executes.

REMAINING KNOWN RED (main-side, routed, NOT this PR's content): with Fix 4 the layering gate finally executes — and returns false on 7 violations that are ALL false edges from #7060's reference-derived producer: import-less std files' prelude bare names (List/fold/concat/Present class) misattributed to extdeps.color.srgb, extdeps.gitignore, v2.compiler.discovery_enumeration. Zero textual imports exist in any flagged file (verified); the files have been import-less for weeks, so this shipped inside #7060 and was never executed post-merge. Routed as dashboard work item adhoc-a7ae8a39-774 (owner sharp-bee-290). Until that lands, batch 4 reds on every PR floor including this one — operator judgment on merging this PR's four fixes on the partial green (batches 1–3 + drift gate) vs waiting.

🤖 Generated with Claude Code