Repair: qualified same-module reference emits wrong module / as fn item

[briansrls]

Auto-opened by session-dashboard for session zesty-carp-424.
Pushing to session/zesty-carp-424 advances this PR.

Worker attestation

• Title describes the change (not the session id or branch).
• PR body summarises what and why (replace the TODO below).
• Tests run: name the command (e.g. npm test, cargo test) and the result.
• If this closes a work item, the body contains a Closes #N directive. (n/a — internal dashboard node, no GitHub issue)
• No commits on this branch are surprises (no fork/cherry-pick I did not make).
• No secrets / credentials / large binaries staged.

Summary

A fully qualified same-module declaration reference used inside a data initializer was emitting against the wrong Rust module, and rendering as a plain function item (foo) instead of a call (foo()). Root cause in emit_var_ref (src/v1/05_emit_rust.dag): a qualified reference (module.leaf) was looked up in the item registry by its full dotted spelling, but the registry is keyed by bare leaf name — so for a same-module qualified reference the lookup missed the intended (correctly-scoped) entry and instead resolved via last-write-wins against whichever module happened to register that leaf name last, and the is_data call-suffix decision fell through to the fn-item (non-call) rendering path. dag/extdeps/container/oci/digest.dag had previously worked around this by respelling the same-module reference as a bare name, leaving a qualified_self_reference_emit_defect_note documenting the dodge.

Fix: emit_var_ref / emit_rust_expr_var now take the current module_name and (1) normalize a self-qualified reference (module_name.leaf where module_name == the emitting module) to its bare spelling before registry lookup, reusing the already-correct bare-name path, and (2) key the general dotted-reference registry lookup on the leaf name rather than the full qualified string, so the is_data call-suffix decision fires correctly for every qualified reference — same-module or cross-module — not just the bare-name case.

The OCI workaround note (qualified_self_reference_emit_defect_note) is deleted: its dissolution trigger — the underlying emitter defect being fixed — has fired. The carrier's spelling in dag/extdeps/container/oci/digest.dag is left as-is (bare reference); only the note documenting the workaround is removed, per the brief (fix the general lowering, not the OCI spelling).

Test plan

New regression witness: dag/test/claim/qualified_declaration_reference_emit_witness_test.dag, with two discriminating fixtures so the fix cannot pass by collapsing both cases together:

• w_same_module_qualified_reference_emits_call — a same-module qualified reference (qdref_same.qdref_same_base) in a data initializer; asserts the emitted Rust contains qdref_same_base() and does NOT contain qdref_same_base.clone() (the fn-item-as-value symptom).
• w_cross_module_qualified_reference_emits_call — a genuine cross-module qualified reference (std.emit_on_demand.interpreter_surviving_roles_ceiling); asserts it still resolves to its real declaring module and renders as interpreter_surviving_roles_ceiling(). This is the sibling that proves the same-module fix didn't just special-case self-references at the expense of general dotted resolution.

Commands run and results:

• cargo run --release -p v1-compiler --bin claim_batch -- --source-root dag --source-root src/v2 --entry dag/test/claim/qualified_declaration_reference_emit_witness_test.dag --functions same_module_qualified_reference_emits_call_not_fn_item,cross_module_qualified_reference_emits_call_from_correct_module — both test fns PASSED by execution (real compile_dag_rust_emit_check compilation against the live corpus, not a mock).
• cargo run --release -p v1-compiler --bin regen_stage0 -- --verify — regen_divergence_count=0, confirmed twice (fixed point: the .dag source and the generated .rs agree).
• cargo fmt --all --check — clean (also enforced by the pre-commit/pre-push hooks on every commit).
• CI run 30761268145 on this PR's head (42ee125) — build, regen, heal_generated_artifacts, and ci (the composed .dag witness/gate floor pass, which discovers and runs the new witness file via the standard dag/test/claim discovery scan) all passed, confirming the fixture executes inside the real regeneration/discovery closure rather than sitting beside it inert.

Not run locally: full cargo test --workspace / cargo clippy --all-targets -- -D warnings. Three genuine attempts (default jobs, reduced jobs, single-crate) all failed with rustc ICEs and sccache spawn failures reproducing even on unrelated trivial upstream crates (unicode-ident, utf8_iter, stable_deref_trait) under heavy host memory/swap pressure — diagnosed as shared-host environment instability, not a regression from this change. CI's build/ci jobs (which do run the equivalent gates on a separate runner) passed, closing this gap.

Scope check against the brief: all six listed items are done — general lowering fixed (not the OCI spelling), same-module + cross-module discriminating fixtures added, correctness proven by execution, fixture sits in the real regen/discovery closure (confirmed via the passing ci CI job), and the OCI workaround note is deleted.