algebra unfork: remove FreeMonoid variant-drop shadow (single authority; §9 narrows scope)

[briansrls]

algebra unfork — remove the FreeMonoid variant-drop shadow (single authority)

Kills the §5 fail-open the de-fork audit tracks as the LIVE {algebra} pair: FreeMonoid/Empty/Cons was defined in both std trees (dag/std/algebra.dag:111 and src/v2/std/algebra.dag), so a closure pulling both silently dropped the coproduct's variant bindings (undefined variable: Empty, reproduced 2026-07-05).

Change (pure v2-tree repoint): delete v2.std.algebra's duplicate FreeMonoid/Empty/Cons, import the single dag/std/algebra authority, repoint the 158 FreeMonoid/Empty/Cons importers. No dag/std edit → no stage0 seed regen; no op moved; no 05_emit_rust.dag (Root-A) touch; no node dependency.

Scope is narrower than the signed design — see docs/plans/algebra-grounding-unification-design.md §9 (execution revision). Three walls, each found by-execution, keep the rest deferred to explicit follow-on lanes:

1. Reserved builtin-method names — a std.* free fn named filter/length/any/contains/skip/is_empty collides with dag/std/algebra's template-declared builtin methods and breaks bare fold(...) resolution corpus-wide. So the operational ops cannot move to a std.* module as-is; they dissolve into the builtin methods in a separate v2 lane.
2. Root-A emitter hardcoding — src/v1/05_emit_rust.dag hardcodes crate::v2_std_algebra::{freemonoid_empty,fold_list,list_snoc_item} for the String=FreeMonoid<Char> host bridge (the design-fenced, unowned Root-A seam). Moving those 3 ops breaks the emitter.
3. QualifiedName is node-gated — its body imports v2.std.node/v2.std.diagnostic/lexing, so promoting it to std.qualified_name needs the separate node defork, not algebra alone.

Verified by-execution: single FreeMonoid definition remains (structural); v1 whole-tree resolve clean (902 sources); the bmc whole-tree v2+dag precompute passes; the generic_alias_coproduct keystone (a both-trees closure over recursive Cons.tail) passes.

[gunbai-bot]

On the failing ci check (floor timeout) — pre-existing, not this change.

The ci job failed as a timeout on the gunbc ci (.dag witnesses + gates) step, not a resolution/logic error. Differential control against main: the last six ci runs on main all fail at 1337–2309s (8254ade, e0da1d4, 293afd2, b209b0e, dfae60e, 21fb5d1) — the chronic corpus-wide O(M²) compile-clean/floor timeout, which is why PRs are being merged manually.

This PR is a pure v2-tree repoint (delete the duplicate FreeMonoid/Empty/Cons from v2.std.algebra, import the single dag/std/algebra authority; net −70 lines, no dag/std edit). It adds no compile burden and cannot have caused a main-wide chronic timeout. Resolution correctness is verified locally by-execution: v1 whole-tree resolve clean (902 sources), bmc whole-tree v2+dag precompute green on the merged head, keystone green. No fix applicable to this check.

— sent from calm-lark-461

[gunbai-bot]

Fixed in 2e4a2a5341. Valid finding — the regex repoint wrongly rewrote import v2.std.algebra inside two String literals that are test input (ts_import_pipeline_source and, same bug, dag_lex_stamp_source), not real imports. Those emit/lex-shape tests parse the string verbatim, never resolve it, and their canonicals expect the 3-segment v2/std/algebra with a manual 3-segment spelling map — so the edit broke the exact-match assertions. Reverted both strings to v2.std.algebra (per your "revert the source string" option — this test is about emit shape, not where FreeMonoid lives). All three affected tests now pass by execution: ts_import_pipeline_exact_holds, ts_import_pipeline_discriminating_red_holds, dag_import_block_lexeme_stamp_holds. A full re-scan confirms no other string-literal imports were touched by the migration.

— sent from calm-lark-461

[gunbai-bot]

ci timeout persists on the current head 2e4a2a5 — still the chronic infra timeout, not this change (updated differential).

The floor step gunbc ci (.dag witnesses + gates) timed out at 10 min again. Fresh differential control against the latest main, including post-#6331 (the floor-resolve speedup that's already in this PR's base): cbf0e38 fails at 973s, 5d1348a8 (#6331 itself) fails at 805s — every recent main run fails. #6331 cut the floor resolve (286s→2.4s), but the compile-clean typecheck is still O(M²) (the type-env-single-authority lever), so the floor times out corpus-wide regardless of the change under test.

This PR is a line-reducing v2-tree repoint; it cannot cause a main-wide chronic timeout. Meanwhile the real code gate rust_tests is green on this head, both api-reviewers APPROVE (no findings), and mergeable=MERGEABLE. No fix applicable to this check on this lane.

— sent from calm-lark-461