Add multi-provider support (OpenAI, Gemini, Mistral/OpenRouter)

[rajkumar42]

Summary

Currently the benchmark runs against a single LLM provider. Add support for running the same benchmark against OpenAI (GPT-4o), Google Gemini (2.5 Pro), and open-source models via Mistral/OpenRouter to enable cross-provider comparison.

What needs to happen

• Add provider configuration (env vars or config file) for OpenAI, Gemini, and Mistral/OpenRouter
• Abstract the LLM call layer so provider can be swapped without changing benchmark logic
• Add a --provider flag or config option to select which provider to run against
• Run the benchmark on each provider and record results
• Document required API keys and setup for each provider in README

Acceptance criteria

• Benchmark can run against at least 3 providers (Claude, OpenAI, Gemini)
• Results are recorded in a comparable format regardless of provider
• README documents how to configure and run with each provider