fix(eval): apr eval no longer reports fake pass@1=1.0 on broken models (PMAT-702)

[noahgift]

Summary

Defect 3 from the PMAT-701 5-whys cascade. `apr eval --task humaneval` was silently reporting `pass@1 = 1.0` (164/164 problems "passed") whenever inference failed — a false positive that masked the Phase 4 Stage D no-KD training run from operators for 2 days.

The bug

When inference failed for ALL samples, the legacy code "fell back to structural validation" and marked every problem with a non-empty `canonical_solution` as `pass=1`. The `mode: "structural"` field was the only signal, easy to miss in automated JSON parsing.

Verified on gx10 against the known-gibberish 10K Stage D checkpoint (PMAT-701 origin):

```json
{
"passed": 164,
"pass_at_k": [{ "k": 1, "rate": 1.0 }, ...],
"mode": "structural"
}
```
exit code: `0`

Fix

`crates/apr-cli/src/commands/eval/inference.rs`:

• `run_humaneval` (the bug site, ~line 134): emit structured `mode: "inference_failed"` JSON with `inference_error` populated and pass counters all zero, then return `CliError::InferenceFailed` (exit code 8).
• `run_mbpp` (~line 1513): MBPP already returned Err on failure but didn't emit JSON. Aligned to the same shape per FT-EVAL-FAILURE-003 parity falsifier.

`crates/apr-cli/src/commands/eval/code_eval.rs:425-428`:

• Removed misleading "falling back to structural validation" log line.

Contract

`contracts/apr-eval-humaneval-inference-failure-handling-v1.yaml` (validates clean via `pv validate`):

• 3 equations: pass@k definition, failure-signal coupling, per-problem counter invariant
• 4 falsifiers: FT-EVAL-FAILURE-001 (broken model→0), 002 (healthy model→real), 003 (humaneval/mbpp parity), 004 (no dataset-side marking)
• 2 Kani harnesses; qa_gate F-EVAL-FAILURE-001

Verification on gx10

`evidence/apr-eval-inference-failure-pmat-702/launch-after-fix.json`:

```json
{
"passed": 0,
"pass_at_k": [{ "k": 1, "rate": 0.0 }, { "k": 10, "rate": 0.0 }, { "k": 100, "rate": 0.0 }],
"mode": "inference_failed",
"inference_error": "No tokenizer found (no embedded tokenizer and no sibling tokenizer.json)"
}
```
exit code: `8` (`CliError::InferenceFailed`)

Cascade context

This closes the third defect from the original PMAT-701 finding. Combined with #1863 (allocator), #1869 (Q4K teacher), and #1871 (spec + dispatch default), the full distill → eval pipeline is now honest end-to-end on Grace Blackwell.

Test plan

• `cargo build --release --features cuda -p apr-cli` — clean
• `cargo fmt -p apr-cli --check` — clean
• `pv validate contracts/apr-eval-humaneval-inference-failure-handling-v1.yaml` — clean
• FT-EVAL-FAILURE-001 verified on gx10 (gibberish checkpoint → `pass@1=0.0`, exit 8)
• CI: `ci / gate` + `workspace-test` green
• Follow-up: run on a known-healthy 7B Q4K teacher to verify FT-EVAL-FAILURE-002 (real `pass@k` still works)

🤖 Generated with Claude Code