Allow users to submit hardware benchmark results to improve model recommendations for everyone.

Requirements

• nexus benchmark runs inference tests on installed Ollama models
• Outputs structured JSON: hardware profile, model, tokens/sec, latency
• nexus benchmark --submit anonymizes and uploads to public dataset
• No PII — hardware profile, model, performance metrics only
• GitHub handle attribution is opt-in
• Submitted data feeds hardware detection recommendations (Hardware detection and model recommendations in TUI #29)

Part of v0.3.0 — Dynamic Routing.