Variant collision diagnostic + factor algebra_templates_for_profile into std/algebra.dag

[briansrls]

Summary

Variant collision diagnostic, algebra enrichment factoring, warning infrastructure removal, review violation fixes, and doc updates.

1. Variant collision diagnostic (Stream 1, violation #1)

Adds a VariantCollision error diagnostic when two explicitly imported enums collide on a variant name. On collision, the ambiguous binding is removed from locals (fail-closed) — later phases see an unresolved variant rather than a fabricated winner.

• 00_core.dag: new VariantCollision { variant, enum1, enum2, span } variant in CompilerDiagnostic
• 04_infer.dag: VariantFoldState accumulator with pre-computed imported_enum_names index (FF-compliant); collision removes binding instead of overwriting
• Stage0 Rust mirrors (v2_std_core.rs, v2_compiler_infer.rs)

2. L1 Tier 2 — factor enrich_kernel_type to call .dag functions in std/algebra.dag

Moves algebra_templates_for_profile from 04_types.dag into std/algebra.dag, co-locating the dispatcher with the data table and per-profile template functions.

• dsl/std/algebra.dag: adds algebra_templates_for_profile function
• src/v2/04_types.dag: imports dispatcher from std.algebra, removes local definition
• v2_compiler_infer_types.rs: adds BooleanAlgebraCollectionProfile variant + templates, fixes Set/NonEmptySet profile mapping

3. Warning infrastructure removal (INVARIANTS.md C-8)

Per INVARIANTS.md: "no warning severity for semantic errors." Promotes all warnings to errors and eliminates the warning concept.

• OwnershipWarning → OwnershipViolation; VariantCollisionWarning → VariantCollision
• is_error_diagnostic → always returns true
• Removed "error"/"warning" severity branching
• dsl/std/symbols.dag: renamed SymbolId variants to fix real collision with SemanticColor

4. Review violation fixes (3 violations resolved)

Violation 1 — No fallbacks that fabricate: On variant collision between two explicitly imported enums, the binding is now removed from locals instead of fabricating a winner. Later phases see an unresolved variant.

Violation 2 — Facts Flow Forward: Pre-computed imported_enum_names as a Map<String, Bool> from resolved_imports before the variant fold. Inner fold performs O(1) map lookups instead of rescanning the import list per variant.

Violation 3 — Early Detection (ownership→emit boundary): Added a gate after ownership analysis in compile.dag and stage0 Rust. If ownership_diagnostics produces errors, the pipeline returns early with empty files instead of proceeding to emit.

5. Documentation updates

• INVARIANTS.md: Marked escape hatch C-8 as FIXED
• ROADMAP.md: Checked off enrich_kernel_type Tier 2, per-profile field builders Tier 1, Set/NonEmptySet profile Tier 2.5
• docs/bootstrap-b-and-lane-work-plan.md: Added DONE status banners to sections 1 and 2

Testing

• cargo clippy --all-targets -- -D warnings — clean
• cargo test --workspace --exclude v2-compiler-tests — 12 passed
• cargo test -p v2-compiler-tests — 282 passed
• cargo test -p v2-compiler-tests full_dsl_compiles -- --ignored — passed
• cargo test -p v2-compiler-tests strict_compile_diagnostic_count -- --ignored — ratchet holds at 0

  

[briansrls]

Review (2 violations, 2 good, 1 improve)

Invariant violations (2)

Modeling (good)

• dsl/std/algebra.dag Moving the profile-to-template dispatch into std.algebra keeps the algebra witness as the single authority for which operations a kernel type inhabits, which is the right compositional layer.
• src/v2/04_types.dag Deleting the local algebra_templates_for_profile helper and importing the algebra authority removes a duplicate representation and improves harmony with the existing kernel_algebra_profile model.

Modeling (improve)

• src/v2/00_core.dag VariantCollisionWarning models namespace ambiguity as advisory lint, but the domain fact is an unresolved ambiguity and would be more faithful as an error or explicit ambiguous-binding state.

The algebra helper extraction is a compositional improvement, but the new variant-collision handling is not fail-closed and introduces a fresh repeated-scan regression.

[briansrls]

Invariant violation: When two explicitly imported enums share a variant name, this code still overwrites locals with one enum after only emitting VariantCollisionWarning, fabricating a winner for an ambiguous reference and violating No fallbacks that fabricate / Explicit boundary contracts.

[briansrls]

Invariant violation: The new prev_is_imported check rescans resolved_imports inside the inner variant fold instead of carrying an indexed import-membership fact forward, violating the Performance invariant's Facts Flow Forward rule.

[briansrls]

Invariant review (6d2716a8)

Violations (2):

• src/v2/04_infer.dag:2524 When two explicitly imported enums share a variant name, this code still overwrites locals with one enum after only emitting VariantCollisionWarning, fabricating a winner for an ambiguous reference and violating No fallbacks that fabricate / Explicit boundary contracts.
• src/v2/04_infer.dag:2516 The new prev_is_imported check rescans resolved_imports inside the inner variant fold instead of carrying an indexed import-membership fact forward, violating the Performance invariant's Facts Flow Forward rule.

Improve:

• src/v2/00_core.dag VariantCollisionWarning models namespace ambiguity as advisory lint, but the domain fact is an unresolved ambiguity and would be more faithful as an error or explicit ambiguous-binding state.

The algebra helper extraction is a compositional improvement, but the new variant-collision handling is not fail-closed and introduces a fresh repeated-scan regression.

[briansrls]

Review (1 violations, 3 good, 3 improve)

Invariant violations (1)

Modeling (good)

• src/v2/00_core.dag Adding VariantCollision to the closed CompilerDiagnostic sum is the right core-layer model because ambiguity becomes a typed compiler fact instead of an ad-hoc message.
• src/v2/04_infer.dag VariantFoldState is a coherent infer-local product type because it threads both chosen locals and accumulated collision diagnostics through one fold state.
• dsl/std/symbols.dag The Status* rename cleanly separates symbol identity from SemanticColor, which is a faithful std-layer distinction.

Modeling (improve)

• src/v2/00_core.dag Now that every CompilerDiagnostic is effectively an error, the boundary would compose more cleanly if severity disappeared structurally instead of being represented by a degenerate classifier.
• src/v2/04_infer.dag A more compositional model would represent ambiguous variant lookup as its own binding shape instead of winner-plus-error, so later phases cannot consume a fabricated enum choice.
• dsl/std/symbols.dag If the symbol vocabulary keeps growing, compose standard_symbols from per-domain tables so the flat list stays derived instead of hand-expanded.

The PR improves the diagnostic and symbol models, but fail-closed handling is still incomplete and the previously flagged variant-collision issues remain unresolved.

[briansrls]

Invariant violation: Stage 6 now emits OwnershipViolation diagnostics here, but the pipeline still proceeds through artifact planning and emit without gating on ownership_diags, so a known error crosses the ownership→emit boundary in violation of the Early Detection invariant.

[briansrls]

Invariant review (0cecc3e0)

Violations (1):

• src/v2/compile.dag:122 Stage 6 now emits OwnershipViolation diagnostics here, but the pipeline still proceeds through artifact planning and emit without gating on ownership_diags, so a known error crosses the ownership→emit boundary in violation of the Early Detection invariant.

Improve:

• src/v2/00_core.dag Now that every CompilerDiagnostic is effectively an error, the boundary would compose more cleanly if severity disappeared structurally instead of being represented by a degenerate classifier.
• src/v2/04_infer.dag A more compositional model would represent ambiguous variant lookup as its own binding shape instead of winner-plus-error, so later phases cannot consume a fabricated enum choice.
• dsl/std/symbols.dag If the symbol vocabulary keeps growing, compose standard_symbols from per-domain tables so the flat list stays derived instead of hand-expanded.

The PR improves the diagnostic and symbol models, but fail-closed handling is still incomplete and the previously flagged variant-collision issues remain unresolved.