refactor(hir): audit #[allow(dead_code)] stubs in cssl-hir

[tmchow]

Summary

Closes #2. Audits the three #[allow(dead_code)] items in cssl-hir against "has the T3.4-phase-2.5 work landed or is it still pending" per the issue's decision tree, and applies the follow-up the issue prescribes for each case.

Outcome per item

1. compiler-rs/crates/cssl-hir/src/cap_check.rs:137 — fn emit(..., span: Span)

Pending. DECISIONS.md § T5-D3 (Cap-check pass sig-level only for stage-0; full expr walk deferred) explicitly defers "full linear-use tracking through every expression" and "handler-one-shot enforcement" to T3.4-phase-2.5. The Session-1 STATUS line lists T3.4-phase-2-refinement / T3.4-phase-3-AD-legality / T3.4-phase-3-IFC as ✓, but not T3.4-phase-2.5.

Per the issue: "If the work is still pending: add a tracking comment with the DECISIONS.md entry it's waiting on." Upgraded the bare // reserved for T3.4-phase-2.5 expression walk to an explicit pointer to § T5-D3 so the next person opening cap_check.rs lands on the actual DECISIONS entry instead of having to grep for T5-D3 themselves. #[allow(dead_code)] kept.

2. compiler-rs/crates/cssl-hir/src/cap_check.rs:231 — fn matrix(...) -> &AliasMatrix

Same dependency as (1). matrix() is the accessor the deferred body walk needs to call AliasMatrix::can_pass_through / param_subtype_check at call sites (per § T5-D3's "ready for use when T3.4-phase-2.5 walks call-args" note). Same tracker comment upgrade.

3. compiler-rs/crates/cssl-hir/src/lower.rs:1271 — const fn _span_referenced(_: Span) {}

Genuinely dead. The stub exists only so Span in the use cssl_ast::{...} line doesn't trip -D warnings. Grepping lower.rs for �Span� returns exactly three hits: the import on line 21, the stub's /// comment on line 1270, and the stub's signature on line 1272. Nothing outside the stub uses Span.

Per the issue's guidance: "If the work is complete and the function is genuinely unused: remove it, or wire it up if it was meant to be called." Removed both the stub and Span from the import. SourceFile stays in the import — it IS used at lines 49, 56, and 919.

Verification

$ cargo build --workspace
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 14.62s

$ cargo test -p cssl-hir
test result: ok. 195 passed; 0 failed; 0 ignored

Clean compile across the workspace (no new unused_imports warning from the Span removal), all 195 cssl-hir tests pass.

Scope note

The issue lists three items. This PR addresses exactly those three — no opportunistic removal of other #[allow(dead_code)]s elsewhere in cssl-hir. If the project wants a broader sweep later, it's a follow-up issue's scope.

Closes #2

---

This contribution was developed with AI assistance (Claude Code).