fix(kernel): fix embedding model config being hardcoded and ignored

[woodcoal]

Problem:

• In kernel.rs, embedding driver creation hardcoded the model name
• Config file setting memory.embedding_model was completely ignored
• Always used "text-embedding-3-small" or "nomic-embed-text" instead of user config

Solution:

• Modified create_embedding_driver calls to use config.memory.embedding_model
• Applied config model in all three auto-detection paths:
  1. When embedding_provider is explicitly configured
  2. When auto-detecting OpenAI
  3. When auto-detecting Ollama
• Added logging to show the actual model being used
• Maintained backward compatibility: use defaults when config is empty

Scope:

• crates/openfang-kernel/src/kernel.rs: lines 757-803

Testing:

1. Set custom embedding_model in config file
2. Check logs after startup to verify model name
3. Verify vector search functionality works correctly