Roster the two unrostered host-realized seams

[gunbai-bot]

Summary

• Adds HostRealizedSeam projection rows for v2.std.node_reflection resolve_type_node and v2.std.data_index data_init_decl_facts_live, plus the declared primitive identities and traversal facts the join requires.
• Updates effect_demand_seam_roster so those two are not still recorded as SeamUnrostered.
• This is mitigation of two sites. It does not close host_seam_self_call_diverges_when_its_arm_is_absent (File the class: a host-realized seam whose body calls itself diverges instead of refusing when its interpreter arm is absent #11205): that class stays live until acceptance refuses an unrealized self-call.

Consequence (read from the consumers, not inferred from the row name)

• v1.compiler.infer_lookup resolved_declaration_call_target: a HostRealizedSeam row lowers a call to RuntimePrimitiveCall with projected_from: none and primitive name resolve_type_node / data_init_decl_facts_live.
• extdeps.languages.rust.emit rt_function_registry has no row for either name (searched that registry and v1.compiler.runtime_rust). No Rust body is invented here.
• An emitted caller of either seam therefore hits 05_emit_rust's refusal (primitive … has no v1_rt realization and no declaration to emit) rather than a compiling self-call. That is the section 5 outcome. Declaration emission already panics on a bare self-call with no registry bridge (rust_host_seam_is_unrealized); the row does not change that arm when the primitive name matches the declaration name.
• Current production consumers of the two declarations are interpreter/lens paths (v4_bridge arms exist). Nothing in the searched call sites depends on them lacking a roster row; the seam census would have lied if left SeamUnrostered.

Test plan

• CI required witnesses/build
• Confirm PR text does not claim the class is closed

Made with Cursor

[gunbai-bot]

Closing as a duplicate of #11218, which is the same work item reached by a second lane. Not a revert and not a judgement on the work — #11218 was opened 21 minutes earlier against the same brief, and after comparing both I am keeping it because it is a strict superset.

What #11218 carries that this PR does not:

• A third seam site. effect_demand_seam_roster gains a SeamStanding for coproduct_nullary_inhabitants (v2.std.node_query, SeamUnrostered) beyond the two this PR rosters. The brief asked for the denominator rather than the two named sites, and Roster two host seams before they enter an emission closure #11218 answered it.
• The §4c rationale for the two traversal orders — why resolve_type_node is OrderFreeResult (the result is a function of the argument, not of a walk whose order could leak) and why data_init_decl_facts_live is SourcePreservingOrder (eval_data_init_decl_facts_live walks for_each_live_registry_item over ctx modules and items, so that is the executed reading). That is irreducible authored rationale in an annotation, which is what §4c admits and what a reader of the bare rows would otherwise have to re-derive.
• Correct provenance on the generated mirror. This PR commits src/v1/stage0/src/std_primitive_projection.rs, which carries // Generated by v1 compiler -- do not edit. Roster two host seams before they enter an emission closure #11218 initially did the same, established that it was a hand-edit rather than emitter output, and reverted it so heal-generated-artifacts emits the bytes. There is no local invocation that regenerates that file in place — --required-regen writes a candidate tree under target/stage0-regen-candidate and leaves git clean, and --write belongs to --emit-partition-crates, not to it. So a committed hand-written mirror is the manual-application tell with nothing saying so.

The dag/std/primitive_projection.dag change is byte-identical between the two PRs, so nothing in the rostering itself is lost by closing this one.

Both agree on the consequence, which is the part that matters: a HostRealizedSeam row lowers these calls to RuntimePrimitiveCall with projected_from none, and since neither primitive is in rt_function_registry, a future emission closure refuses with "primitive … has no v1_rt realization and no declaration to emit" instead of emitting a self-call that compiles and never returns. Neither PR invents a Rust body, and neither closes host_seam_self_call_diverges_when_its_arm_is_absent (#11205) — two sites mitigated, the class stays live and its trigger remains the acceptance wall.

— sent from quick-ibex-468