Route-A final-six: clear the E0308 family to 0 (milestone, not cargo-green)

[briansrls]

Route-A final-six: clear the E0308 family to 0 (a milestone, not cargo-green)

Clears the 6 remaining out-of-scope E0308 in the faithful --emit-fresh seed crate to 0, via three emitter-faithfulness fixes in src/v1/05_emit_rust.dag (spliced into the committed seed v1_compiler_emit_rust.rs via the 2-emit bootstrap). Each fix is faithful construction derived from the declared type, with a per-family discriminating witness + negative control. Zero unwrap_or_default.

Honest scope — read this first

• This clears the E0308 grep-family to 0. It is NOT cargo-green. The faithful crate still has ~32 other-family pre-existing errors (measure-tower E0560×12 + E0614×9, E0425×5, E0573×2, E0599×2, E0091×2, E0107/E0277/E0063×1). cargo-green = 0 total cargo errors; E0308 is one grep-subset.
• These 32 are pre-existing, not introduced or unmasked by this PR. Constant across the lane: fresh-C 41 = 9 E0308 + 32 → fresh-E 38 = 6 + 32 → fresh-G 32 = 0 + 32. The 6-tail touches none of the 32 (identical-32 control → regression-free).
• Next step (separate, gated): a per-error emitter-vs-model diagnosis of the measure-tower before any fix — std.measure is load-bearing and the magnitude grounding is an open DESIGN thread, so it does not get a cargo-green-chasing patch here.

The three fixes

1. &str borrow seam (emit_typed_bin_op): a String + String concat now borrows the RHS (l + &r), gated on op_str == "+" && both_string; optional-string compares wrap Some(..).as_deref() both sides. Witnesses: string_concat_borrows_rhs (+) / numeric_plus_not_borrowed (control: numeric + is untouched — the borrow is string-typed-derived, not blanket).
2. Box/Rc (emit_data_def_body BoundedLattice special path): top/bottom are bare generic T, so they are emitted bare, not Box::new'd (meet/join fn-fields still get Rc::new). Witness bounded_lattice_top_bottom_are_bare_not_boxed asserts no top:/bottom: Box::new( and meet: Rc::new( present (proves only the wrong Box-wrap was removed, not all field wrapping).
   • Two edits, one effective: the real fix is the emit_data_def_body special path. The companion removal of the dead wrap_rust_record_field_value BoundedLattice branch (old 6142-6143) is a proven no-op §3 cleanup — the corpus has zero inline BoundedLattice record-literals (all 3 are data-defs), so that branch never fired.
3. fn-item single-field un-collapse (emit_typed_record_lit): a single-field record whose field is fn-typed must un-collapse to Rc::new(R { field: Rc::new(fn) }) — the bare collapse would drop both the nominal type and the fn→Rc<dyn Fn> coercion (a fn-item where Rc<dyn Fn> is wanted → E0308). This is the data semver_scheme: VersionScheme = { compare: semver_identity_compare } site.
   • Fail-closed recovery (the load-bearing crux): a NEW helper find_unique_struct_name_by_fields returns the struct name only when exactly one struct matches the field set (count == 1), else None. It does not reuse find_struct_name_by_fields, which returns candidates |> first (pick-first). On ambiguity the fix falls back to bare collapse — a loud typed E0308 at the consumer — and never guesses a nominal R. "Pick the first candidate to make it compile" is exactly the forbidden silent-wrong-nominal-type.
   • Three discriminating evidences (route_a_final_six_test.rs):
     • Effectiveness — fn_field_single_record_uncollapses_with_rc_wrap: a unique single fn-field record un-collapses to Scheme { pick: Rc::new(cmp) } (asserted on the signature-stripped body, so Scheme { proves the struct literal, not the return type).
     • Transparency — nonfn_unique_single_field_stays_collapsed: a unique non-fn single field (recovered unambiguously) stays collapsed — proves the un-collapse is gated on fn-ness, not merely on a unique nominal recovery.
     • Fail-closed-on-ambiguity — ambiguous_single_field_name_fails_closed_to_collapse: two fn-field structs sharing field name pick → count == 2 → None → bare collapse, never emits AScheme {/BScheme {.

Flagged (not fixed here)

• Latent fail-open in find_struct_name_by_fields (multi-field): it still picks the first candidate on ambiguity. Not exercised by this PR's single-field path, but it is the same pick-first silent-wrong-nominal hazard one rank up — flagged for a follow-up, not papered over.
• The deeper Optional/Value::Null model-forces-unwrap follow-up stays operator-upstream (the open ==/Value::Null-split DESIGN thread); this PR does not expand into it.

🤖 Generated with Claude Code