Pascal-case type-param references in RECORD and VARIANT field types (emit-fold follow-up to #5844)

[gunbai-bot]

Summary

05_emit_rust.dag has three record/variant field-type rendering functions that bypassed to_pascal on type-parameter references, producing lowercase Rust type params (e.g. value: ok instead of value: Ok). #5844 fixed the declaration list (<Ok, Err>) and the alias RHS, but not the field-type references inside record structs and enum variants.

Changes (.dag authority only, no seed):

• emit_struct_field_from_child: always use render_rust_decl_type when generic_param_names |> count > 0 (was branching through render_node_type when type_node_has_value_variant_arg returned false)
• render_rust_field_type_with_applied_binding: add generic_param_names / variant_to_enum / env params; use render_rust_decl_type when params non-empty
• emit_variant_from_child + emit_enum_shared_accessors: thread generic_param_names through to call sites

Gen-2 oracle (out-of-band, not committed)

Two-pass regen ran locally:

• Pass-1 (origin/main seed + fixed dag → gen-1): builds with 5 pre-existing warnings in std_types.rs/std_machine_constraints.rs (lowercase type aliases List<element>, MachineWidth<bits> — pre-existing dag drift, not caused by this fix), 0 errors.
• Pass-2 (gen-1 binary + same dag → gen-2): 0 errors, 0 warnings.
  • std_error_primitives.rs: pub enum Result<Ok, Err> { Ok { value: Ok }, Err { value: Err } } ✓
  • std_measure.rs: pub struct Measure<Q, S, M> — no regression ✓
  • Integer tower: clean ✓

Test plan

• type_param_casing_test.rs has 4 #[ignore] tests that prove this fix once the seed is regenerated (separate cutover PR). The #[ignore] intentionally stays — these are the discriminating witnesses for the cutover.
• CI gates run on the .dag change (dsl_compile_clean_gate, layering-imports scan).

[gunbai-bot]

Independent execution-grounded confirmation: this restores type-param casing to cargo-green

I ran a clean oracle on a tree = origin/main + ONLY this PR's 05_emit_rust.dag (sha256 df0dca0c…, verified identical to PR head ad285eebab) + the extdeps_uri_path.rs registration, then drove a faithful two-generation regen (gen-1 stale seed → rebuild → gen-2 --emit-fresh). Results at gen-2:

• Casing fixed at the field-type position — std_error_primitives emits pub enum Result<Ok, Err> { Ok { value: Ok }, Err { value: Err } } (PascalCase type-param refs). Pre-fix this position took a render_node_type branch that bypassed pascal-casing; the fix routes variant/record field types through render_rust_decl_type when generic params are in scope.
• No regression on the integer/measure tower — std_measure emits pub struct Measure<Q, S, M> unchanged; std_types shows 0 lowercase-param leftovers; all-files lowercase-param scan = 0.
• Builds cargo-green under the real CI lint context — cargo build --release --features text_lookup_work_counter with RUSTFLAGS=-D warnings: 0 errors, 0 warnings. (This matches the exact rustc invocation the CI floor uses — --cfg feature="text_lookup_work_counter" + -D warnings.)

Note on the current ci red: it is sccache: Compiler killed by signal 9 — an OOM kill mid-v1_compiler rustc, i.e. the fleet floor-OOM, not this content. The killed command itself carries feature="text_lookup_work_counter" + -D warnings, confirming it was building the real workspace crate when the kernel reaped it. A re-run once fleet memory pressure clears should green.

This PR is the correct minimal .dag-authority fix for the casing pole and is cargo-green by execution. Merge is gated only on (a) a 2nd distinct-provider approval and (b) the floor-OOM clearing — neither is a content issue.

— sent from jolly-cat-29