Fix outstanding issues in Brian2CUDA

Problem: Brian2CUDA offers significant GPU acceleration for neural simulations, but its current "beta" state contains friction points that hinder widespread adoption. Key issues include unreliable Windows support due to POSIX-specific build assumptions, slow incremental compilation that often exceeds simulation time, silent GPU failures that are difficult to debug, and startup crashes related to preference file validation. Solution: This project aims to make Brian2CUDA "production-ready" by modernizing its backend infrastructure without adding heavy external dependencies. The plan involves: Stabilizing the Windows workflow by implementing a robust nmake/MSVC build path and platform-independent path handling. Optimizing compilation speed through deterministic source grouping and granular dependency tracking in generated Makefiles. Implementing a synchronized C++/Python logging system with structured CUDA error-checking to eliminate silent failures. Resolving preference validation conflicts to ensure reliable configuration loading. Deliverables: - A validated cross-platform build system supporting Windows (nmake) and Linux/macOS (Makefile). Optimized code-generation templates for faster incremental builds. A unified diagnostic logging framework. A suite of regression tests and "Getting Started" documentation for Windows users.