Upgrade GEPA from 0.0.22 to 0.1.4

[Superagentic-AI]

Summary

• Bump gepa from 0.0.22 to 0.1.4 (PyPI latest) in pyproject.toml and refresh uv.lock.
• Keep CodeOptiX on the existing minimal integration (InstructionProposalSignature via MinimalGEPAProposer); no public evaluation/evolution API changes.
• Document GEPA 0.1.x prompt-template placeholder rename: <curr_instructions> / <inputs_outputs_feedback> → <curr_param> / <side_info>. Default templates come from GEPA, so callers that do not pass a custom prompt_template keep working.
• Update the GEPA LLM wrapper to accept str | list[dict] prompts (LanguageModel protocol in 0.1.x). Text-only reflection remains the common path.
• Changelog entry under Unreleased; codeoptix package version left at 0.1.3 (dependency bump only; public CodeOptiX APIs unchanged).

Compatibility notes (SuperQode / soft imports)

• Soft-imported surfaces under evaluation/evolution/behaviors/adapters/Bloom/CLI are unchanged.
• Residual risk: any external config that sets a custom prompt_template with the old 0.0.x placeholders will raise ValueError until updated to <curr_param> and <side_info>.

Test plan

• uv lock resolves gepa==0.1.4
• uv pip install -e ".[dev]"
• Smoke: MinimalGEPAProposer.propose_improved_prompt with fake LLM against gepa 0.1.4
• pytest tests/ -v --cov=src/codeoptix → 86 passed
• ruff format --check src/ → pass
• ruff check src/ → same 12 pre-existing findings as on main (PLR0917/UP042/PLC0207); no new issues in touched files