Move the bottom seam to std.error_primitives (re-derives #13399)

[gunbai-bot]

Re-derives the seam of closed #13399 on current main; the old head is history, not a base.

What

pure_dag_seam_unreachable and its _float / _string projections leave std.bytes for std.error_primitives (the leaf: imports nothing, already imported by std.algebra). In std.bytes the seam sat above std.types -> std.algebra, so std.algebra could not name it and trim spelled bottom as an unnamed from_code_point(1 / 0).

Delete-first: the three fns are removed from std.bytes (nothing re-exported them), then every importer re-pointed: std.encoding, std.decimal, gunbc.auth.approval_device_wire, gunbc.auth.oidc_id_token_verification, extdeps.languages.gpu.wgsl.kernel, extdeps.languages.simd.kernel, extdeps.cloud.gcp.secret_manager (enumerated by grep over dag/ and src/; none outside dag/). std.algebra trim now diverges through the named seam. Seed mirrors (v1 maintenance, serves the v2 self-host): std.primitive_projection row module_path, runtime/v1_rt/interpreter panic text, and the generated v1_compiler_runtime_rust.rs text.

Not carried over

The from_code_point / std.unicode.scalar / recurring-failure-mode / src/v2 witness edits in #13399: they belonged to the stacked #13381, not to the seam.

Evidence

• claim_batch on a seed built from this branch, dag/test/claim/error_primitives_bottom_seam_test.dag: PASS bottom_seam_projections_resolve_from_error_primitives, PASS trim_still_routes_to_the_host_realization. A bad import path (std.bytes) refuses at resolve, which is the discriminating red.
• Instruments (container; srv1 unreachable, systemd-run not installed here), seed built from this tree's v1 sources: gunbc test //gunbc/instruments:self-host exit 0 and gunbc test //gunbc/instruments:v2-native-cli exit 0. (Run on the tree stacked with item 2, whose diff touches no src/v1; origin/main baseline self-host also exit 0.)

Correction to the first push

The first head broke the required gate: std.error_primitives sits below std.primitive_projection in the import graph (projection -> decl_ref -> types -> algebra -> error_primitives), so no closure of it contains the roster, its self-call never resolves to the bridge, and the emitted crate failed function cannot return without recursing @ std_error_primitives.rs. Root fix at the emitter's owner: emit_fn_def (src/v1/05_emit_rust.dag) emits a NULLARY realized seam as its own v1_rt::<bridge>() call (it already reads the roster directly via rust_host_seam_is_realized; a nullary seam has no by-ref/wrap choice). Stage0 mirrors regenerated to first_generation_equal=true. Parameterised seams still route through call-target resolution unchanged.

🤖 Generated with Claude Code

Refresh 1667581 (2026-10-11): merged origin/main (merge commit); stage0 mirrors regenerated, never taking either side: round 3 reported first_generation_equal=true. self-host exit 0, v2-native-cli exit 0 on this head (container run; srv1 unreachable, no systemd-run).

[gunbai-bot]

Instrument receipts (run in the session container, which has no systemd-run; cgroup memory.max is 24 GiB and the gunbc test memory envelope was enforced by the runner): gunbc test //gunbc/instruments:self-host exit=0; gunbc test //gunbc/instruments:v2-native-cli exit=0, on seed built from head bc22614. — sent from jolly-ibex-327