fix(llm): preserve reasoning content across turns

[atirna]

Summary

• preserve LiteLLM reasoning_content on assistant messages that stay in the conversation history for a later model call
• carry the field through the standard agent, assist, interact, MCP, and crew paths while omitting it when a provider does not return it
• Fixes Bug: DeepSeek‑V4‑Flash with thinking mode enabled throws reasoning_content must be passed back to the API in multi‑turn tool‑call workflow #96

Validation

• pytest -q tests/unit/llm/test_reasoning_content.py tests/unit/llm/test_api_base.py tests/integration/test_agent_loop.py (38 passed)
• Black and Ruff checks on the changed files
• python -m compileall -q pentestagent/llm/llm.py pentestagent/agents/base_agent.py pentestagent/agents/crew/orchestrator.py
• git diff --check