Stage0 mirrors reconciled with their .dag authorities (zero-drift regen); release-locus constants registered

[gunbai-bot]

Brings main's stage0 seed mirrors into line with their .dag authorities, so the next PR that regenerates a mirror stops tripping over drift that was already on main (#11985, #12047 and #12070 each had to hand-pick one line).

Receipt: fresh regen round on this head

The binary was built from this head in a private CARGO_TARGET_DIR:

required-regen: elapsed_ms=563935 first_generation_equal=true planned=159 executed=159 adjudicated=159 declared_divergent=1 [main.rs]
required-regen: first_generation_equal=true candidate=target/stage0-regen-candidate

• The round did not refuse.
• Every committed mirror is byte-identical to its candidate (checked with cmp, not only by the drift list).
• generated_artifact_gate main (dry) exits 0.

Per-mirror disposition

Mirror	Disposition
release_locus_seed_constants_generated.rs (round refused: "committed mirror is no longer emitted")	The authority is intact; its registration was missing. The file's producer is still gunbc.release_locus_seed_constants_emit via generated_artifact_gate main_wet, the same wet-actuator path as gunbc_file_transport_generated. #12011 never added its WetActuatorGeneratedRegistration row, so compile_stage0()'s population reported it as lost. Fix: add the row (PlainPubMod) in v2.compiler.self_host.stage0_crate_layout, then regenerate bootstrap_stage0_crate_layout_generated.rs and dag/gunbc/stage0/stage0_crate_layout_generated.dag through main_wet_one, then regenerate the gunbc_stage0_crate_layout_generated.rs mirror. lib.rs is unchanged: after the rebuild the candidate emits its pub mod line itself.
emitted_population.rs	(a) Mirror was stale (two emitted paths missing). Installed the regenerated file.
std_integer.rs, std_machine_constraints.rs	(a) Mirror was stale. The .dag declares WidthResolution as a coproduct (StaticWidthIndex | PointerWidth); the mirror had a hand-written PointerWidth struct and type alias in its place, which is the E0573 residue. Installed the regenerated files.
v1_compiler_emit_rust.rs	(c) The three re-exports second, NonEmptyStr and literal_suffix have no consumer through v1_compiler_emit_rust. A search found none, and the crate builds without them. Dropped.
v1_compiler_infer_resolve.rs	(a) The "TRANSIENT BOOTSTRAP PATCH" was the hand-carried projection of an arm v1.compiler.infer_resolve already declares. The regenerated file carries that arm, plus the later kind_inhabitant_matches_resolved arm the .dag has since gained, so no hand change is lost. Installed the regenerated file.
v1_compiler_emit.rs, v1_compiler_infer.rs, v1_rt.rs	These do not drift at ee04f784624. The candidate is byte-identical to the committed file, so the v1_rt content-hash change is already carried by its authority (landed via #12065). No change.

No seed Rust was added, so no seed-retained row was needed. No CI job was added.

Notes

• --regen-round-cost refused on this tree with convergence kind projection returned unknown closed variant Refused and installed nothing. The loop was driven by hand instead: --required-regen, install, rebuild, repeat. This is a separate defect and is not fixed here.
• A round run on a binary that fails to build is not a receipt. One intermediate round printed first_generation_equal=true over a lib.rs that did not compile, because the emitter was a stale binary. The receipt above comes from a binary built from this head.

🤖 Generated with Claude Code