Performance: Add lazy loading for rarely used generated models

[knowlen]

Description

As identified in PR #32 review, the large number of generated models could benefit from lazy loading to reduce memory usage, especially for users who only use a subset of the API.

Current State

• All generated models are imported at module initialization
• Memory footprint includes all models even if unused
• Impacts startup time and memory usage

Proposed Solution

Implement lazy loading using import hooks and a registry pattern:

class LazyModelLoader:
    def __init__(self):
        self._models = {}
        self._loaded = set()
    
    def __getattr__(self, name):
        if name not in self._loaded:
            # Dynamically import the model
            module = importlib.import_module(f'esologs._generated.{name}')
            self._models[name] = module
            self._loaded.add(name)
        return self._models[name]

# Usage
models = LazyModelLoader()
# Model only loaded when accessed
character_data = models.get_character_by_id

Implementation Ideas

1. Use Python's import hooks for transparent lazy loading
2. Track which models are commonly used together
3. Provide option to preload common models
4. Consider using __init__.py with __getattr__ for module-level lazy loading

Benefits

• Reduced memory footprint for applications using subset of API
• Faster initial import time
• No change to public API

Performance Metrics to Track

• Import time before/after
• Memory usage for minimal usage patterns
• Memory usage for full API usage

Considerations

• Ensure thread safety for lazy loading
• Consider impact on type checking and IDE support
• May need to update documentation about import behavior

References

• PR Release v0.2.0b1 - First Beta Release #32 review feedback
• Python import hooks documentation