refactor(llm): Clean up __init__.py - move logic to dedicated modules

[m96-chan]

Problem

src/pygpukit/llm/__init__.py is 688 lines with too much logic that should be in dedicated modules.

Current State

src/pygpukit/llm/
└── __init__.py  (688 lines - imports + logic)

Issues

• __init__.py should primarily be exports
• Contains initialization logic that belongs elsewhere
• Hard to understand module structure
• Circular import risks

Proposed Changes

1. Move model loading logic to loader.py
2. Move configuration to config.py
3. Keep only exports in __init__.py
4. Target: < 100 lines in __init__.py

Proposed Structure

# __init__.py (< 100 lines)
from pygpukit.llm.models import QwenModel, LlamaModel
from pygpukit.llm.loader import load_model, from_safetensors
from pygpukit.llm.config import ModelConfig
from pygpukit.llm.chat import ChatSession

__all__ = [...]

Related

• refactor(llm): Split model.py by model architecture #141 (model.py split)
• refactor(llm): Split layers.py by layer type #142 (layers.py split)

[m96-chan]

Completed in PR #168