[codex] upgrade GEPA and add observability benchmark

[ASRagab]

Summary

• Upgrade GEPA to 0.1.1 and constrain LiteLLM to versions past the affected supply-chain window.
• Make optimizer evaluator execution parallel by default while preserving serial mode with --no-parallel and documenting worker tuning.
• Add the optimization observability benchmark workflow, scorer scripts, strict report checks, OpenSpec specs/archive entries, and tests.

Validation

• git diff --check
• uv run python scripts/check.py --skip-smoke
  • 326 passed, 8 skipped
  • score checks passed for skills/generate-evaluator/SKILL.md, evaluator-cookbook.md, and skills/optimization-guide/SKILL.md

Notes

• Draft PR because this branch includes a dependency lockfile update and new benchmark workflow that may deserve maintainer review before marking ready.