Add all remaining operator guides and complete README docs hub

[sunway513]

Summary

• Add 6 new operator guides covering every operator in AITER
• Update README.md with a complete documentation hub linking all 9 guides
• Every operator in the Supported Operators table now links to its relevant guide

New Documentation (this PR)

GEMM Variants & Tuning Guide

• A8W8, A16W16, A4W4, batched, DeepGEMM, Triton FFN fusions
• Complete tuning system docs (CSV format, env vars, model configs)

Quantization & Precision Guide

• QuantType enum, per-tensor/token/block strategies
• Fused quantization ops (FP8 + MXFP4), SmoothQuant, KV cache quant

Normalization Guide

• RMSNorm, LayerNorm, GroupNorm with all fused variants
• Add + SmoothQuant + Dynamic Quant fusions, backend dispatch logic
• Fused QK norm + RoPE + cache + quant mega-kernels
• Distributed RS + RMSNorm + Quant + AG fusion

RoPE (Rotary Position Embedding) Guide

• SBHD, THD, 2D, 3D tensor formats
• NeoX & GPT-J rotation styles, partial RoPE (nope_first)
• 8 scaling methods (Linear, NTK, YaRN, Phi-3, DeepSeek, LLaMA3, MRoPE, DualChunk)
• Autograd classes for training, fused operations

KV-Cache Management Guide

• Paged, flash, ASM, and MLA cache layouts
• Quantized cache (FP8, INT8, FP4) with per-token and per-block scales
• Fused RoPE + cache write, fused BMM + RoPE + cache
• Block swap/copy for beam search and speculative decoding

Elementwise & Activation Guide

• SiLU/GELU/sigmoid/tanh activations
• SwiGLU/GeGLU gating (the standard LLM FFN pattern)
• Fused activation + quantize (FP8 group, MXFP4 block-scale)
• Binary arithmetic with broadcasting, fused mul-add

README Updates

• Documentation table expanded from 5 to 9 guides
• All 13 operators in the table now link to their guide

Test plan

• Verify all doc links in README resolve correctly
• Review each guide for accuracy against source code
• Confirm formatting renders properly on GitHub

🤖 Generated with Claude Code