style: silence the dormant graph-retention seam #1648 left red on main

[justinchuby]

origin/main (73e6fe15a) fails a required check on an unmodified checkout:

error: methods `set_retain_decode_graph_across_spec` and
       `retain_decode_graph_across_spec` are never used
  --> crates/onnx-genai-engine/src/native_decode/cuda.rs:5516

That is the Check the native backend compiles step of Rust quality, so it blocks every open PR regardless of contents. It surfaced on an unrelated CPU-kernel PR (#1628).

Both methods are #[cfg(test)] accessors for the option-c graph-retention seam that #1648 landed as an enabling primitive — deliberately ahead of the WP4 tests that will drive them. Their #[cfg(test)] siblings either side (set_retain_graph_on_rewind, padded_query_capacity) are already called, which is why only these two trip.

Fix is #[allow(dead_code)] on the pair with the reason recorded at the site — rather than deleting a seam that is about to be used, or widening the allow to the whole impl block.

Local verification

gate	result
cargo fmt --all -- --check	clean
clippy -p onnx-genai-engine --features native-backend --all-targets -D warnings	clean
clippy -p onnx-genai-engine --features native-cuda --all-targets -D warnings	clean
clippy -p onnx-genai-cli --all-targets -D warnings	clean
cargo test -p onnx-genai-engine --features native-backend	0 failed

Process note — this is the third main-is-red repair today, and I am only finding them because they land on an unrelated PR:

• this morning: fmt + clippy debt from Native MTP self-spec: land Gap 2 (hidden-seed shape) + Gap 1 wiring; block on int4 shared LM-head #1637/feat(mtp): on-GPU int4 draft LM-head projection + native MTP E2E wiring (recurrent-Scan blocker) #1641 (four required gates red at once) → fixed by fix(engine): make main green — fmt, dead code, and three 1.98.0 lints from #1637/#1641 #1640
• midday: fix(mtp): correct recurrent-verify oracle; prove MTP already token-identical (real blocker = graph reuse) #1644 reintroduced fmt violations → fixed by style: rustfmt the two sites #1644 left red on main #1642
• now: perf(mtp): add second CUDA graph slot for option-c verify capture (enabling primitive) #1648 introduces this dead-code lint

Each costs a full CI round-trip to discover, because the gates are sequential — the clippy steps only run once formatting passes. Running the Rust quality job on main post-merge, or a merge queue, would catch these at the source instead of on whoever's PR is open next.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 81.40%. Comparing base (73e6fe1) to head (c17b155).
⚠️ Report is 3 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1649      +/-   ##
==========================================
- Coverage   81.46%   81.40%   -0.07%     
==========================================
  Files         384      382       -2     
  Lines      180299   177247    -3052     
  Branches   180299   177247    -3052     
==========================================
- Hits       146886   144292    -2594     
+ Misses      28467    28025     -442     
+ Partials     4946     4930      -16     

Flag	Coverage Δ
cli-ort-linux	82.60% <ø> (ø)
cli-ort-windows	82.11% <ø> (ø)
mlas	?
offline	81.36% <ø> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.
see 3 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.