Parser decodes: route NotUnicodeScalar into each parser's own typed refusal (XL-2 from_code_point; lands inside #13378's unit)

[gunbai-bot]

Stacked on #13378 (bright-fox-380, still open). This branch merges session/bright-fox-380-fcp, so the diff against main includes #13378 until it lands. That PR declares std.unicode.scalar from_code_point / char_text and files RFM bare_from_code_point_binds_the_total_seed_builtin.

Lands as part of one unit (manager decision): #13378 deletes the seed from_code_point builtin. The seed interpreter dispatches builtins before module functions (v1_interpreter eval_call: eval_builtin before ctx.lookup_fn), so until then every call here still binds the total builtin, and the witnesses below cannot go green on this head alone. This PR is approved on its own diff. bright-fox-380 then merges it into #13378 together with #13386 / #13384 / #13381, and one gate runs over the combined head.

This is XL-2's from_code_point follow-up for the parser-decode population. It is PR 1 of 2 in this lane.

Survey: re-deriving the 9 PARSER DECODE identities

identity	finding	disposition here
extdeps.languages.yaml.ingest::yaml_double_quoted_escape	a \x/\u/\U escape decoded from source	migrated: NotUnicodeScalar → YamlScalarRefused, which names the escape and the cause (YAML 1.2.2 §5.1 and §5.7). The hand-duplicated surrogate/range check is deleted
tools.fabric_ci_evidence::fabric_ci_decode_step	a 6-hex-digit word decoded from the wire	migrated: a new typed and located refusal FabricCiEvidenceUndecodable { cause: FabricCiDecodeRefusal }. Its arms are FabricCiWireUnframed, FabricCiWireEmpty, FabricCiWordNotHex { at } and FabricCiWordNotScalar { at, cause: NotUnicodeScalar }. The hand range check is deleted; 8 consumer patterns are updated in gunbc.fabric_control_plane_live_probe and tools.fabric_ci_evidence
extdeps.languages.json.parse::json_unescape_decode_piece	the whole .dag json_unescape chain had zero consumers. parse_json decodes through the native json_unescape_checked, which recombines surrogate pairs and refuses unpaired ones (RFC 8259 §7)	deleted (json_unescape, json_unescape_decoded_pieces, json_unescape_decode_rest, json_unescape_drop_first, json_unescape_decode_piece, plus their acc types). The existing RED is test.claim.json_parse_witness::an_unpaired_surrogate_refuses_rather_than_vanishing
std.judgment_contract::string_from_code_points	input is chars(s) of a String, so it is CHAR-kind and not parser decode; no refusal is possible	deleted. length_prefixed_decode now uses the declared std.coercion unicode_scalar_unfold / unicode_scalar_fold, and the dead re-export in std.materialization_provider is removed
extdeps.standards.rfc_8949::cbor_text_of_octets, extdeps.http.form_urlencoded::form_component, gunbc.auth.approval_device_wire::decode_path_segment, gunbc.auth.oidc_id_token_verification::jwt_segment_json, extdeps.uri::uri_percent_decode_component	under-typed UTF-8. These get code points that utf8_decode_octets already validated as scalars but returns as List<Int>; uri hand-rolls a second decoder. A NotUnicodeScalar RED is unauthorable at the call	PR 2 (manager ruling): the decoder yields Char, the callers use char_text, and uri is routed through the one std decoder

Module-grain consequence (manager ruling, migrate-all): importing from_code_point into yaml.ingest rebinds every bare caller in that module. So its 6 CHAR-kind sites move here too, and the CHAR lane drops the module:
yaml_refused_characters, yaml_first_refused_character, yaml_line_start_mark, yaml_line_join_mark, yaml_key_separator, ingest_yaml_source. The refused repertoire is now one literal yaml_refused_text whose chars are the List<Char>, spelled through char_text, and the marks are "\u{N}" literals.

Migrated / deleted identities (several are outside the required gate)

• extdeps.languages.yaml.ingest: yaml_double_quoted_escape, yaml_double_quoted_scalar (new), yaml_first_refused (new), yaml_refused_text (new), yaml_refused_code_points, yaml_refused_characters, yaml_first_refused_character, yaml_line_start_mark, yaml_line_join_mark, yaml_key_separator, ingest_yaml_source
• tools.fabric_ci_evidence: FabricCiEvidenceDecode, FabricCiDecodeRefusal (new), FabricCiDecodeState, fabric_ci_decode_step, decode_fabric_ci_evidence_wire, and two consumer matches
• gunbc.fabric_control_plane_live_probe: 6 FabricCiEvidenceUndecodable matches
• extdeps.languages.json.parse: the json_unescape chain (deleted)
• std.judgment_contract: string_from_code_points (deleted), length_prefixed_decode
• std.materialization_provider: re-export removed
• RFM bare_from_code_point_binds_the_total_seed_builtin: receipts updated. PARSER DECODE is re-derived to 5 remaining and re-kinded; CHAR goes from 85 to 79

REDs (one per parser)

• yaml: test.claim.yaml_ingest_witness::red_a_non_scalar_hex_escape_refuses_typed_and_located. \uD800 refuses at line 2 naming a surrogate, \U00110000 refuses as past U+10FFFF, and the U+D7FF / U+10FFFF neighbours decode
• fabric: test.claim.fabric_ci_evidence_wire_decode_witness::red_a_non_scalar_word_refuses_typed_at_its_offset. The real-path control is a_wire_the_encoder_wrote_decodes_back, which round-trips through the real encoder
• json: the existing an_unpaired_surrogate_refuses_rather_than_vanishing

Also: json_hex_nibble had two importers reaching it through a re-export in json.parse. They now import it from its declarer, extdeps.languages.json.grammar: extdeps.bmc.ipmi_master_write_read and gunbc.machine_intake_ampere_socket_console_observation.

No rows were added to floor_cross_claim_pure_producers_warm.

🤖 Generated with Claude Code

[gunbai-bot]

Follow-up on review 76486: I merged #13378's current head (merge commit 48adbb9c75). It now includes the CHAR (#13386), octet and seam lanes, so the CHAR population named in the review (std.content_hash::content_hash_combine_preimage, extdeps.dns.domain_name::fold_dns_case_string_at, …) is migrated on this head.

This PR is part of the #13378 landing unit. The builtin deletion and all five migrations land together, as gunbc.recurring_failure_mode bare_from_code_point_binds_the_total_seed_builtin (DISPOSITION OF THE SEED BUILTIN) records, and this PR never lands alone.

The reviewer's generated-Rust point stands as a gate item for the combined head: the emitted rt_function_registry row (extdeps.languages.rust.emit) and the primitive-surface / std.primitives from_code_point_contract rows must agree with the deleted builtin before the unit lands. The remaining bare callers on this head are the five byte-input sites migrated further up this lane's stack (#13389 / #13390) and test.claim.git_upstream_model_witness (octet lane); the combined gate's grep is where 0 is asserted.

— sent from bold-deer-208

[gunbai-bot]

Closing as merged into #13378: this branch's head eea9a7e is an ancestor of #13378's branch (session/bright-fox-380-fcp), verified with git merge-base --is-ancestor. Its content lands as part of #13378's from_code_point unit (the seed builtin deletion plus all bare-caller migrations, landing together), so it never lands alone. Its review approvals apply to this content inside #13378.

— sent from lively-crane-656