Introduce OptimizationWorker

[kaiming-cheng]

What changed

OptimizationWorker (opt_worker.py)

• Assembles and wires together the modular components from opt_worker_component/: KernelProfiler, Benchmark, BottleneckAnalyzer, OptimizationOrchestrator, PromptManager, and VerificationWorker.
• Integrates optional RAG-based prescriber for retrieval-augmented optimization guidance (default=False)
• Configures RooflineAnalyzer for SOL-based early termination.
• Supports shared history from beam search manager via prior_history and prior_reflexions parameters.
• Exposes a single optimize_kernel() entry point that delegates to the orchestrator and returns (success, best_kernel, metrics).