test(codex): add automatic recall regression benchmark

[AirHua-byte]

Description

Adds a privacy-safe benchmark for the real Codex automatic-recall hook. It records latency, false-injection rate, positive-recall rate, and injected-token estimates, and compares reports produced from the same fixture without publishing query text, URIs, or recalled content.

Human Involvement

• A human participated in the implementation or review loop
• This PR was generated entirely by AI agents without human participation in the loop

Related Issue

Part of #3686

Type of Change

• Documentation update
• Test update

Changes Made

• Exercise the real Codex auto-recall hook against JSONL fixtures.
• Emit aggregate latency, recall, false-injection, and token-budget metrics only.
• Add same-fixture report comparison and a synthetic public negative fixture.

Testing

• Added focused tests
• Targeted tests pass locally
• Tested on macOS

PYTHONPATH=. <validated-python> -m pytest -o addopts='' --confcutdir=tests -q tests/benchmark/test_codex_recall_benchmark.py — 5 passed; 4 existing dependency deprecation warnings.

Checklist

• Code follows the project's style
• Self-review completed
• Documentation updated
• The tracked fixture contains synthetic negative prompts only

Additional Notes

Deployment-specific positive fixtures stay local and untracked. Reports contain no query text, URI, or recalled memory content.