feat: add Claude Code CLI as LLM backend

[themoddedcube]

Summary

• Adds ClaudeCodeLLM, a new LLM provider that uses the Claude Code CLI (claude -p) as a subprocess for generation
• No API keys needed — authentication uses the CLI's existing OAuth session (claude login)
• Adds a provider field to LLMModelConfig and a provider registry in LLMEnsemble for automatic backend selection
• Includes 9 unit tests covering initialization, CLI invocation, system message passing, error handling, and provider registry integration

Usage

Config-based (no code changes needed):

llm:
  provider: "claude_code"
  models:
    - name: "sonnet"
      weight: 1.0
      max_tokens: 16000
      timeout: 300

Programmatic (for custom setups):

from openevolve.llm.claude_code import init_claude_code_client

for model_cfg in config.llm.models:
    model_cfg.init_client = init_claude_code_client

Files changed

File	Change
openevolve/llm/claude_code.py	New ClaudeCodeLLM class + init_claude_code_client factory
openevolve/llm/__init__.py	Export new classes
openevolve/llm/ensemble.py	Add provider registry + _create_model() dispatch
openevolve/config.py	Add provider field to LLMModelConfig
tests/test_claude_code_llm.py	9 unit tests

Test plan

• All 9 new unit tests pass
• All 376 existing tests still pass (pre-existing failures unrelated: missing requests module, integration tests needing API keys)
• Manual verification with claude CLI installed and authenticated

🤖 Generated with Claude Code

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}