v0.4.1

What's New

• force_conversion parameter — convert() now accepts force_conversion=True to run the full source→IR→target pipeline even when source and target providers match, ensuring parameter normalization (e.g. max_tokens → max_completion_tokens for OpenAI Chat)
• Vendored validate.py updated — reimported from zerodep v0.4.1 with type introspection caching (functools.cache) for faster validation

Fixes

• Remove incorrect "Related Projects" section from README
• Apply ruff format to benchmark scripts
• Remove unused sys import in benchmark script

Benchmarks

• Add LiteLLM vs LLM-Rosetta conversion comparison benchmark
• Add real-payload benchmarks using production-scale payloads (31-41 tools, up to 218 messages)