feat: Add KV-Cache Manager for LLM inference (#223)

## Implementation

User-space KV-cache management system for LLM inference optimization.

### Features
- POSIX shared memory pools for cache storage
- Multiple eviction policies (LRU, LFU, TTL, Size-based)
- Persistent cache with save/restore
- Thread-safe operations
- CLI interface for cache management
- Comprehensive test suite

### Files
- kv_cache_manager.py: Core implementation
- test_kv_cache_manager.py: Test suite

### Usage
```bash
cortex cache create llama-cache --size 16G --tier cpu
cortex cache status llama-cache
cortex cache persist llama-cache
cortex cache restore llama-cache
cortex cache evict llama-cache --percent 25
```

Closes #223

Author: yaya1738