Remove rust_tests job from CI (free a runner slot; drop the 37 GiB job)

[gunbai-bot]

What

Removes the rust_tests CI job entirely (operator directive: free runner capacity). The v1 fmt+nextest gate was non-required and red on main (it tests the v1 seed, which is being deleted per §7), and at ~37 GiB it was CI's single largest memory job — so removing it frees a runner slot and the fleet's biggest memory consumer. Formatting stays enforced by the pre-push cargo fmt hook.

Done as a complete removal, not a half-measure: rather than leave orphaned coproduct variants (which a completeness witness would flag red — the tolerated-red anti-pattern), the removal is clean through the model.

Changes

• ci.yml: rust_tests job gone (−107 lines); regenerated via main_wet, drift-clean. CI is now ci + ci_regen.
• Gate / surface coproducts: removed the now-orphan RustMonolithGate gate and GithubActionsRustTestsJob surface variants and every match arm (ci_gate, ci_floor_plan, merge_admission, floor_materialization, commit_workflow), the roster enrollment + projection, and the dead gate-runner wrappers (ci_gates, floor_effect_gate_witness).
• Materialization: re-homed the fresh-effect cgroup-measurement from the retired rust_tests step to the ci-job floor peak reads; bumped the step-count pins (uses 9→6, run-natures 20→14) and the placement-grain duplicate-computation flagship to the surviving 2 sibling jobs (still >1, still discharged by sccache).
• Witnesses: updated realization-binding count (3→2) and a drive-by fix of a pre-existing stale golden (affected set processing #6453 added floor_skip_discovery_witness to the build-verify list but left this unenrolled witness's golden stale).

Also includes docs/plans/space-lens-minimal-project.md (a plan doc already on this branch).

Verification (green by execution)

main_wet regen passes (proves the coproduct removal resolves in the emit closure). Ran the affected witnesses locally: ci_spec_witnesses, commit_workflow_witnesses, ci_floor_plan_witnesses, floor_materialization, the ci_materialization count pins (6/14/≥2), and the placement-grain flagship (ci_build_across_two_jobs_is_one_violation) — all PASS, including witness_gate_roster_matches_coproduct_arms (confirms every remaining Gate variant is still enrolled).

Follow-up

The now-dead dag/tools/rust_gates_ci.dag tool module is orphaned (no importers) and can be deleted in a follow-up.

🤖 Generated with Claude Code

[gunbai-bot]

Thanks — the coverage-drop concern is valid and I've addressed it with option (b); one factual correction on the framing.

Correction — it was not a required gate. On the gunbc main ruleset, only the ci job is protection-required; rust_tests (and ci_regen) are not protection-required, and rust_tests was in fact red on main (it exercises the v1 seed, which is being deleted under §7). So this isn't a required-merge-protection regression — it's the removal of a non-required, already-red gate. That's precisely why the operator ruled to delete it rather than keep tolerating a permanent non-required red (§5), and it was the fleet's single largest CI memory job at ~37 GiB.

Why not (a) re-enroll on the ci floor: that would move the ~37 GiB red job into the required gate — the opposite of the intent, and it would make the required ci job inherit the v1 seed's redness. Explicitly rejected.

Fix — option (b), pushed in 78bc01b: a declared Terminal disposition (commit_gate_rust_suite_removed_disposition) at commit_gate_roster — the discoverable home for "why is there no rust gate?" — recording the removal as intentional with its rationale and its no-return trigger: the suite does not re-enroll; it is retired with the v1 seed and deleted at §7 terminal collapse. Coverage after removal is stated explicitly: cargo fmt stays enforced by the pre-push hook (CommitCargoFmtCheck), cargo nextest run -p v1-compiler-tests runs locally, and compile-level seed breakage is still caught by the required floor's DagCompileCleanGate + RegenVerifyGate — only the seed's runtime test behavior moves local.

— sent from merry-owl-649