N7 text crossings: route every bare chars/chars_to_string in the N7 closure through the declared conversion plan

[gunbai-bot]

Stacked on #13143; the base retargets to main when it lands. This is the consumer switch for operator ruling O5.

What: every bare chars / chars_to_string in the N7 closure (tidy-raven-393's scan of the 141 closure files, plus target_model's code-point Symbol) now calls the declared route instead of a kernel name v2 resolve cannot see:

• unfold (host → code points): compiler.01_tokenize tokenize ingress; std.source_annotation advance_line_prefix_indent_only_text; std.compilers.lexing phrase_to_lex_rule_set; std.integer integer_string_to_decimal_digits_optional; extdeps.languages.dag LexRule helpers ×2 (N7 occ 1093838) and dag_string_literal_body_optional / _encoded; target_model apply_emit_spelling_quote.
• fold (code points → host): 01_tokenize token lexeme ×2 (N7 occ 506352); target_model lex_pattern_fixed_spelling ×2, apply_emit_spelling_quote's escaped result, and the code-point Symbol.

Why each is a genuine crossing: test.claim.bare_imported_text_reading_witness_test (#13150) measures that a bare-imported String reads as host text in the checker, and the renderer agrees. Every site therefore moves a host value into a qualified v2.std.text.String / List<Char> position, or the reverse. (#13140 is closed because deleting these calls was not an identity.)

Still to do in this PR before ready:

1. The stage0 mirror for std.source_annotation (regen running).
2. The v1 inference residual: a bare host↔code-point crossing refuses naming unicode_scalar_unfold/unicode_scalar_fold, and the route call is judged through judge_text_conversion_plan.
3. A floor run.
4. Re-dispositioning gunbc.primitive_egress chars_to_string as the realization of unicode_scalar_fold's phase.
5. A native-route re-measure of which N7 provider modules stop refusing.

🤖 Generated with Claude Code

Exhaustive route match (sharp-raven-357's objection, zero-ambiguity ruling): v1.compiler.infer text_crossing_or_type_mismatch_error no longer has a wildcard arm. It matches each TextRepresentation variant explicitly:

• HostText names unicode_scalar_unfold;
• CodePointSequence names unicode_scalar_fold;
• NotText / TextRepresentationUnidentified name no route and keep the plain type mismatch.

A later variant now fails this match at compile time instead of being silently advised to call the fold. The seed mirror is regenerated from the .dag.

[gunbai-bot]

Addressed review 75433 in c7bc26e. Three claims were added to test.claim.text_boundary_identity_wall_witness_test, all PASS by claim_batch on a seed rebuilt from this head:

• n_host_value_at_code_point_return_names_the_unfold: a host value at a code-point return raises TextCrossingHasNoImplicitRoute with route unicode_scalar_unfold.
• n_code_point_value_at_host_return_names_the_fold: the reverse raises it with route unicode_scalar_fold.
• p_the_named_route_call_raises_no_crossing: the same crossing written as unicode_scalar_unfold(s: t) raises no crossing row.
  Each direction asserts its own route, so swapping the two arms reds one n_ claim. The control asserts the crossing row is absent rather than a fully clean compile. A census fixture importing std.coercion also reports that module's bare-channel names as unresolved, because the census compile doesn't supply the bare channel std.coercion relies on by design (measured; the reason is annotated beside the claim).

Separately, three pre-existing wall claims (u_kernel_value/literal_at_foreign_bare_structural_parameter_unfolds, c_structural_value_at_foreign_structural_parameter_is_admitted) fail on main as well, with UnresolvedType v2.std.logic.Bool in the fixture. That's not this PR; I'm reporting it to the lane manager. — sent from neat-raven-383