chore: remove unused workspace dependency edges

[e54-bot]

Summary

• remove unused wright-language -> wright-core
• remove unused wright-ostw -> wright-ir and dev-dependency edges
• update Cargo.lock

Verification

• cargo fmt --all -- --check
• cargo clippy --workspace --all-targets --all-features --locked -- -D warnings
• cargo test --workspace --all-targets --all-features --locked
• independent metadata/source audit and dependency-edge ablation

Closes #266

[Teakowa]

Blocking — manifest dependency removals (crates/wright-language/Cargo.toml / crates/wright-ostw/Cargo.toml) change an existing compatibility contract. On CI run #440, the OPY N-level gate regresses from 6/6 to 5/6: real-world/overpy-cake still compiles with exit 0 but diverges after normalization. The same gate passed on #260 with the same Rust 1.98.1 runner/toolchain, and only release metadata/docs changed between #260 and this PR's base. This violates #266's requirement to preserve behavior and rule out Cargo feature/unification or other indirect dependency effects. Identify the removed edge(s) causing the divergence, retain any edge that is actually required, and rerun the applicable gate to green.

[e54-bot]

Follow-up on the blocking review: I audited the reported N-level regression and reran the gate. The PR diff is limited to the two manifests and Cargo.lock. Comparing the cargo build -p wright-cli normal/build/feature graph between origin/main and this PR shows only the removed wright-ostw -> wright-ir edge; wright-ir remains directly reachable through wright-driver and wright-transform, while wright-language -> wright-core and all wright-ostw dev-dependencies are outside this production graph. Source/build-script/test audit found no consumer for the removed declarations. With Rust 1.98.1, the PR head passes the N-level gate locally at 6/6; repeated overpy-cake runs are stable, and the base and PR binaries produce identical raw output (SHA-256 f9f4c0930cdf3ff5ba5c7ebd075b6a2ce4e08cf83716cea6e9e733d4754c3b03). The hosted Linux job was rerun on the same commit and still reports 5/6, so I have not retained an unused dependency as a Cargo-graph workaround. A concrete Linux output diff or additional evidence identifying a required edge is needed before changing the bounded cleanup.

[Teakowa]

LGTM