v0.1.45: Fix CrewAI cost enrichment

Fixed

• CrewAI cost enrichment - CrewAI crew execution spans now include gen_ai.request.model, gen_ai.usage.{prompt,completion,total}_tokens, and gen_ai.usage.cost.{total,prompt,completion} attributes
• CrewAI token usage extraction - _extract_usage() now correctly handles CrewOutput.token_usage (UsageMetrics pydantic model), includes cached_prompt_tokens for Anthropic cache cost calculation

Added

• _extract_model_from_crew() helper - Extracts LLM model name from Crew agents' llm attribute (string or BaseLLM object), with fallback to manager_agent