row 155 B2: the protocol seam names no http type; head fields and u16 statuses, the kernel converts at its edges (TODO 34c (a))

[MattJackson]

What

Row 155 "B2", TODO row 34c drain (a): crates/busbar-contract/src/protocol.rs names no http type. BUSBAR-1.6.0.md Part 2 #49 ("core names zero plane types") and Part 4 (every ABI shape lives in busbar-contract and is neutral). Design approved by the ARCHITECT as briefed.

The seam

Before	After
ClaimsFn = fn(&http::HeaderMap, &str)	fn(HeadFields<'_>, &str); claims fns read through the new protocol::head_field (first field whose value is a legal header value, name matched in any case), which is what the struck header_map + HeaderMap::get/contains_key read
DialectCodec::attach_error_response_headers(&mut http::HeaderMap, ..) plus a defaulted error_response_fields	attach_error_response_headers struck; error_response_fields is the required method
EgressAuthHeaders -> Vec<(HeaderName, HeaderValue)>	-> Vec<(String, Vec<u8>)>
quota_exceeded_status: http::StatusCode, auth_failure_status_and_kind: (http::StatusCode, &str)	u16, (u16, &str)
private header_map (HeadFields -> HeaderMap)	struck

codec.rs's WireBody and the pub use http re-export stay.

Callers moved (the kernel converts at its edges)

• busbar-kernel/src/proto/installed.rs detect_protocol: the HeaderMap goes to the claims fold as field pairs, in the map's order (a name's first value first), through ProtocolDecl::claim_over.
• busbar-kernel/src/proxy/mod.rs: ingress_error inserts error_response_fields pairs (the writers insert unique names, so order and multiplicity are the same); auth_failure_status_and_kind converts the u16 (same 401 fallback).
• busbar-kernel/src/ingress/mod.rs limit_refusal: the two quota_exceeded_status reads convert the u16 (same 429 fallback).
• busbar-kernel/src/bound_credential.rs DeclaredBuilder::headers_for: converts the plane's pairs to headers.
• busbar-plane-llm/src/codec/proto_codec.rs DialectRef::error_response_fields: builds its writer's header map on the plane side. ProtocolWriter::attach_error_response_headers and the writers are untouched.
• busbar-plane-llm/src/codec/{anthropic,bedrock,cohere,gemini,openai_chat,openai_responses}/mod.rs: claims signatures and bodies, and the two status fields as .as_u16(). Bedrock's SigV4 sniff keeps to_str's visible-ASCII condition (is_ascii on a legal value).
• busbar-plane-llm/src/exchange/reply/failure.rs: reads the u16 directly.
• busbar-voice/src/lib.rs (decl statuses, voice_egress_auth_headers) and tests/egress_tests.rs.
• Kernel test decls: src/tests/proto.rs, src/proto/tests/neutral_protocols.rs.

Customer-visible behaviour is identical to 1.5.5: header order, multiplicity, case and statuses are unchanged.

RED

crates/busbar-contract/tests/protocol_names_no_plane_type.rs, committed first (e2b7051): a source scan of protocol.rs for http:: / use http. Against predev it lists the use http::{..} line, ClaimsFn, attach_error_response_headers, EgressAuthHeaders, both status fields and the two named() status values. The fix commit adds a_head_field_is_the_first_legal_value_under_its_name_in_any_case (case-insensitive name, a control-byte value passed over, tab legal, DEL illegal).

Overlap check (queue rule 20, widened)

git merge-tree --write-tree of this head against origin/predev 41fd7c5 and each overlapping PR head:

• llm codec: translated bytes back to 1.5.5 (row-159 L3#5-#9, L2#1) #857 9d86513, $ llm codec: reasoning asks and hosted tools stop crossing into Bedrock/Anthropic, as in 1.5.5 #850 d902275, $ llm codec: Cohere TIMEOUT and Bedrock malformed_* stops are served and billed as in 1.5.5 #844 cd9d676, $ llm codec: cache-write and Cohere cached tokens bill at the input tier, as in 1.5.5 #843 f8e5490 (plane-llm codec mod.rs, writers, proto_codec.rs): clean; this PR's hunks are the claims fns and the two DECL status rows in mod.rs, and the DialectRef impl in proto_codec.rs.
• row 130 E5a: every framed connector dial honours the proxy environment, as 1.5.5 did #804 65f28a0, row 130 E6: the plugins.fetch download leaves busbar-kernel and rides the one connector (stacked on #804) #805 b511baf (kernel proxy/mod.rs): clean; this PR edits auth_failure_status_and_kind and the tail of ingress_error.
• row 74 F25: a base URL's or token endpoint's userinfo rides as Authorization: Basic where 1.5.5 sent it; no plane receives it #913 39a1c0f, row 74 F25: the door path's jwt-bearer mint keeps 1.5.5's raw aud, the stripped dial and the Basic #922 0f7a5d8 (bound_credential.rs): clean; this PR edits DeclaredBuilder::headers_for only.
• FLIP-STREAMING: the streaming plane served through its door; busbar-voice deleted #503 f05708d: modify/delete on crates/busbar-voice/src/lib.rs and src/tests/egress_tests.rs. FLIP-STREAMING: the streaming plane served through its door; busbar-voice deleted #503 deletes busbar-voice; predev's voice crate has to compile against the new field types, so the edit can't be avoided. Whichever lands second keeps the deletion.

None of those PRs adds a new use of the moved shapes (their diffs were scanned for the seam names).

Not run locally

cargo build/test/clippy and xtask were not run on the laptop (queue rule 3). cargo fmt was run on the four touched crates. ModernCI is the proof.

[MattJackson]

#1028 (row 155, codec host services) is stacked on this PR: it threads host: &dyn CodecHost through the DialectCodec::error_response_fields hunk in crates/busbar-contract/src/protocol.rs, plus kernel proxy/mod.rs ingress_error, neutral_protocols.rs and proto_codec.rs DialectRef. It stays DRAFT until this lands.

[MattJackson]

Rule 20 note from #1030 (row 155): disjoint hunk in crates/busbar-kernel/src/proto/installed.rs (arm_host_services drops the two usage-tap installs and its doc). A test merge is clean.

[MattJackson]

Rule 20 note from #1031 (row 155, draft): disjoint hunks in installed.rs (arm_host_services drops the cap reader), gemini/mod.rs, openai_responses/mod.rs and proto_codec.rs (the writer's translate_cap). A test merge is clean.