feat: auto-migration for vector dimensions and optimizations for 8GB RAM pc

- Implemented synchronous vector dimension mismatch detection and auto-migration
- Added dynamic dimension probing for Ollama models
- Expanded model dimension registry (Qwen3, DeepSeek, Snowflake, Nomic V2)
- Optimized config.yaml for 8GB RAM systems (Qwen2.5-Coder 0.5B + Qwen3-Embed 0.6B)
- Added regression tests for dimension migration logic

Author: razvan