Add Learn AI Visually to CUDA Learning resources - #4

Adds one bullet under Learning Resources → CUDA Learning.

Learn AI Visually is a browser-based learning site that visualizes the GPU/CUDA execution model — warps, SMs, memory hierarchy, roofline model, coalesced access, tiling, tensor cores, operator fusion / FlashAttention, and Triton vs CUDA — through interactive simulations. Each topic includes a step-by-step explanation paired with a manipulable visualization. No GPU required, runs entirely in the browser, free.

Track outline:

• GPU & CUDA (9 modules): Why GPUs → Execution Model → Memory Hierarchy → Roofline → Memory Access Patterns → Tiling & Matmul → Tensor Cores → Operator Fusion → Triton & torch.compile
• LLM Internals (9 modules) and LLM Serving (7 modules) cover the application layer (KV cache, paged-attention, speculative decoding, multi-LoRA, prefix caching).

Fits alongside the existing CUDA Learning entries (NVIDIA docs, CUDA samples, Programming-in-Parallel-with-CUDA examples) as an introductory visual companion.