Add DeepTempo Direct-Edit + GPT-5-Codex location oracle result

[samarmstrong]

This PR submits a DeepTempo Location Oracle result for Direct-Edit + GPT-5-Codex.

Result summary:

• Setting: PatchEval Location Oracle, comparable to the Location Oracle leaderboard tab.
• Entry type: LLM / non-agent direct edit harness.
• Model: gpt-5-codex via OpenRouter.
• Strict PoC+Unit: 77/230 = 33.48%.
• PoC-only: 82/230 = 35.65%.
• Returned API cost: approximately $0.181/CVE.

Method disclosure:

• Direct search/replace edits -> git diff.
• Used the official problem statement and oracle vulnerable file content.
• One compiler-feedback repair round for compiler/build failures.
• No PoC or unit-test feedback was used during generation.

Operational notes:

• Running the upstream harness locally on macOS/Colima required Docker/bind-mount fixes so /workspace/fix.patch stayed available after container startup.
• The submitted patches are plain diffs and do not depend on those local harness changes.
• Positive controls with embedded reference patches passed 3/3 strict locally.
• Reproduction instructions are included in 2026-07-09__direct-edit-gpt-5-codex__location-oracle/REPRODUCTION.md.