refactor(llm): Split layers.py by layer type

[m96-chan]

Problem

src/pygpukit/llm/layers.py is 1491 lines with all layer implementations.

Current State

src/pygpukit/llm/
└── layers.py  (1491 lines - all layers)

Proposed Structure

src/pygpukit/llm/layers/
├── __init__.py          (exports)
├── base.py              (BaseLayer)
├── attention.py         (Attention, GQA, MHA)
├── mlp.py               (MLP, GatedMLP, SwiGLU)
├── norm.py              (RMSNorm, LayerNorm wrappers)
├── embedding.py         (TokenEmbedding, PositionEmbedding)
└── block.py             (TransformerBlock)

Benefits

• Each layer type in its own file
• Easier to understand layer implementations
• Clear dependencies between layers
• Easier testing per layer type

Related

• refactor(llm): Split model.py by model architecture #141 (model.py split)

[m96-chan]

Completed in PR #168